diff --git a/.gitattributes b/.gitattributes index 544d081a78ad123b38d5a839e55218868eebbd40..387f9bd3abecefa9557cd149ce68d58c2d70d583 100644 --- a/.gitattributes +++ b/.gitattributes @@ -5068,4 +5068,5 @@ lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample9-layer4 lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample94-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample87-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample93-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text -lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample96-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text \ No newline at end of file +lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample96-layer4-item1.zst filter=lfs diff=lfs merge=lfs -textlambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample80-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample162-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample76-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample76-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..f69671a9585b4410479d588f059d6663b80f0fa0 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample76-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1bac700a0db9475e9d55432a5a632ee8a3d128161ed76c34029147b4aeb43ea +size 2471972 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample77-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample77-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..e195885a82ca41fcc9a19ca05b53074a954f3142 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample77-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23f0ea122f1c285db18be7ca40b150f9802baa0f58f3bcc6f625f93a2a35fdd8 +size 2414953 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample78-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample78-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..693ee2b1a2c434748fce7e2f2d22a9a08b9ead19 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample78-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80ea16ca4ce7eb53a36934df466c37259907e939c520ef37d0a26a42342728bc +size 2465191 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample79-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample79-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..4d17cfd316e28d7fe54dc2022966eb71305bfbd3 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample79-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:694f58390776f9b63627a3e9b3b6cbac095981a5e83d83695cf646fec25b9114 +size 2341706 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample8-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample8-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..d6cb95a2b3899f5596373c3a1c508b8825a3e383 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample8-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53bfce0de5912d426e274c8864a37a4ffca33cab8cbc7dd11205f6cdf5cf683a +size 2551470 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample80-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample80-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..ac3f0cd4ac0895721cf9bca8f4beca82336191f1 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample80-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13b4e1fed57b33279746e0a8d9b64979f4439741333f1c94516016247ea84fab +size 2439288 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample81-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample81-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..e8f7bf31ac28261c02e128e1e0e07abe7e52bdf2 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample81-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64df240f08d0a41c5a6bde2933294cc51b70f5fb78db984875ee0775777c47b9 +size 2405096 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample82-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample82-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..d634984c0ad8209a29af78988f4c8862c821106c --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample82-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fefdd9289654d710cdb85c57452ba0a1fa5b08a7ff43c53353e76280a1d7b186 +size 2615382 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample83-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample83-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..cc46857119f77269eb59feb5b558d1de491c3234 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample83-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d261088f9afbf391a79d8a3ece831a00ea831266b6ee6281a387e7546ecf902 +size 2536797 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample85-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample85-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..e75a41963e047befc8d304af415aa68aaff50ba9 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample85-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f42b795a499ed021d4a6fce886ee826b3717353f92aa81b157bf5fe052555e5b +size 2411867 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample86-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample86-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..b8100acdfd5cd18376f2b0b52f86eb4acab27cb8 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample86-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5bccbda9128c304531ba85886014c6a4a6426a49aaee05a810dc6009532ddd54 +size 2428520 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample87-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample87-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..9b381e5e64a6f40cfa6e4ebfa6e50a27113f2cf3 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample87-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6820c478bff814369f63a7977f40aee196b30ddd7ce3e176fc4c3b428caa532 +size 2542021 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample88-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample88-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..6911926cd019b2cddf49de9180d55fa8b731ff40 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample88-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09553343c7355896e3565f75ffc43a2d1af19cc4a322df3c399c911c6a87ce0e +size 2457901 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample89-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample89-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..95fdd4979be7119765bed1edb7c8b16bd3ee17e5 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample89-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:388fc86aa8d8e6c986c61cee11f6ac0a2c3a5d4ec38013577c7f238f2381d77b +size 2424185 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample9-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample9-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..bb83e5031ae2b4e1b5a4e57f70112e82ca14f5b7 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample9-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0fce91496b0cf8322e334e2c3ced1adea68f78722072c11f20224a8500e903d +size 2800408 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample94-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample94-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..98e990cac3bc87cbf1aea8be1ef656fe15746195 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample94-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9fe3cdb0ccfb11cb62dab76863fc612654b45e89ddbfdf4d614859cf5951ee8b +size 2443766 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample95-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample95-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..1e2bc5c9746263a5ee8e829a2932bdb6e3ea67b8 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample95-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b3d6542be0274002d8d295191ffcb3b991d3b8ffc9588296f5613539854beb2 +size 2463415 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample96-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample96-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..cc28432b17ad2d5eb4550f7b377f5a6ececbb804 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample96-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf8ebba48618198d58f5770fd81e48e10abd5eecf240b6fa0262c2c13e68d60d +size 2455745 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample97-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample97-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..c540492d04dfd1758fbac4e34c6d1f1465882f36 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample97-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:55bcd1aa4c55aa1292e78c5c87d33321a406d3fff967771671303818a28287c8 +size 2421040 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample99-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample99-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..2454f1609cee6905bb9c079bb97efd7640236c98 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_hellaswag/sample99-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c6609cd567ca02fe9568bcb4c0393ff0f27fe296fb98465678ec5cd23f0cbb2 +size 2425607 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample0-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample0-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..1ed4f50cafb444783ebc43ef21bc4ebfced2100e --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample0-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ecdd9dc26835f381f850c41098f2886f68e5156de383dfcc4ba49474fff4dd6 +size 2132043 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample1-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample1-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..cb570573c2f23b5c373ac75f7684ab13bf8e80ed --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample1-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2246e8c7b7061e9ce32fea806e041598be59082a108cda2d580293460fd573a3 +size 2138893 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample10-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample10-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..e5d8d0c22c550227ed9da87e037938f6fa562651 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample10-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7f37f92d9e6313e7656ae59727b1c7435294c424672a8fc31266197c3301f72 +size 2153124 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample101-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample101-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..767110e29e671ad1e8f84ec328f7a61fce1f9bc4 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample101-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0ca427adc921dd353d6291c5d4490d89674cb9e6756498b7715ecca4c2f38c3 +size 1655709 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample102-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample102-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..93dbc42921ec962aa509af0910bd6d26a438da3b --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample102-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5f67e6fd1d14bdbb56618579f72a5a473e04467ddc94c3d37e723171a90906b +size 1642086 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample106-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample106-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..e3e7665b6d475302cf72fa5d6a0841619e6e5dba --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample106-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f80cb5b2ba937c85e752090a4c63830c972a6b04116161bd7ddf38e16f42e06 +size 1760329 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample108-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample108-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..359f964586cf97ac10712b8073e7bfdfe064c927 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample108-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea8972ab2215f9c7e71f4d866f4fd3a813f2d7228452761ff06916140f95500f +size 1629062 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample110-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample110-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..706b7ed68081a290ccf778dc7304a386fe86dc66 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample110-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7999bb220ffb9ed03656f0e9ad7866f39bd974b789c4b0207d5b8df9c2c0ffe6 +size 1821472 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample115-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample115-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..e0360085174cb3c28932584e60d766cc677e641d --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample115-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b65ed49a3a987ee135137957fd2707b3831de01106df41c95f205ad784f851c +size 1664662 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample116-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample116-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..9353bf2523d771009cbd816e7f2a4ae5858f818b --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample116-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d82137f528926a637d57e46b14bdcbfd5f542c3e015d923525248a01c3b4ff2e +size 1736569 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample12-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample12-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..62e622691531bb578be5d1e39eb031415cdbcaa3 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample12-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b11ab355bd3463d568e21bd9b76ade1d14694559c887342a64a2690b5e6f38d +size 1927807 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample120-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample120-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..e3c7d5b8f0ee6d12cfbe851980ec5914f47d110f --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample120-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f05e4e534159c807d5ff4c421bf802bf086e2aa0608077fa85330d02ba238fc5 +size 1722459 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample122-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample122-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..9fdfb0e0f5de05445987a155ce24ac7f28a87b56 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample122-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3bce3d6342718e143dee0a7c946e848e1e9747dde683d088cf29d2dc0447e1f1 +size 1798164 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample127-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample127-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..c631ce06e716421b392031606435f03f2470c542 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample127-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:929229d2c12ee9b21e0fff8d2aa789bce6ae49553c8a41ab15edb7de939b50b4 +size 1697261 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample128-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample128-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..a3a6f96256ab37913beedec5a2044f5ec879dc18 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample128-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:51217afad790be36453397bb02d81e0c2a3688b2c26d541e44f1c08c89a0cc6e +size 1663943 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample129-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample129-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..3e74abbd9872325a28a375d870fc1992b6fadeb5 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample129-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:17a41f060400de87f128949cfbf20b6f8b6a264fb6c4a9a01b9edd7b8b4b825f +size 1720476 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample130-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample130-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..221bc18f10e347c1696e17a8e87e805c843eef3d --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample130-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc1737afa8fc4e9165a81bf6f1766b76d375aa4432c777bc12387c4661703a8f +size 1655122 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample134-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample134-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..13797686d9d49a4030f6663b8bce9a29d085d0f4 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample134-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:47821b1bc03db76dc6ca629048534f0109ee1cfa6f0eec54e9c9b7c12216b22c +size 1740258 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample135-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample135-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..5d987f83c904e4218e04ebd8cb33cecb3aed5736 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample135-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1637bcaa76a86dd6d47d97709666a6366177e403406660c47c20e5fdd48b556d +size 1718948 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample142-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample142-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..95a29339cfe34ae5bbe269f7b4cee2fb59dcb916 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample142-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:79cfc7d7348a1c914987fdddd9f3e60533f1ae47a7dbaa90167f6b5470007234 +size 1652801 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample143-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample143-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..afe93a89a7ad3ffac8baba05c1f72dea2e8fcaec --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample143-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7116d809d8c5101ba1a2ab776419a34355df90e835670705c21c253d2e4bd549 +size 1605916 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample144-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample144-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..b4afcf448db999e21fb04214dea605c9def90927 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample144-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8e765d5af4e73c7f5ade7c11e27f949d842e1fa4969e5ad7628d3d064978880e +size 1655396 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample148-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample148-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..7f18e0b7bbeec53f55891b6b557f00497d4bd554 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample148-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1b682a73ca31496df2737b3db5693f2c7ae60884bc954b1e24d42187323204c +size 1814777 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample149-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample149-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..ecbfc64d828ee4b7ecc79e0078cf1a80e45154f2 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample149-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1cbccd0ce13ac0d4b75cfae4c3439661066abe5280e3cf17f8a8c96ade4332b8 +size 1658913 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample153-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample153-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..20d161a9ef8a0f6e651b290d3f776adc8548ab79 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample153-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a374c0b41993398f41ac998c93ad531b51147a986dd612b4acc8f078d19e9ebe +size 1642946 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample156-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample156-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..282ea63745b24e12b3fa10ea698b2a12adbda55e --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample156-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4c951d71ed4fdfed2aca99fb94f2b73609d370c37dfbe722f6ed240f09994b3 +size 1635331 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample158-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample158-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..64962c3906eaf493aae503c2892698c87db94ee2 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample158-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83c6b59542480e42c1fdd097089932eff339e84e1bf3356d7b11c03f70a53739 +size 1633645 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample159-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample159-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..154babb78e642742a0c41502f18d97ac43991494 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample159-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00f4c003bad068767697c3fe13ed3705d91b405882031f07114f01393be57707 +size 1630009 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample162-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample162-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..887260b78f0f14c7e42768f4d840bf9f11438264 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample162-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2fe86820f14491ab570677691defdae9ef6362e54b1bce46ca618e50e4b87f91 +size 1597896 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample166-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample166-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..d5ea89fd527528d79b6a654ca659fb996dd47528 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample166-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0eba19236e3a07c94e15d4f1728fe828ba92b63b0fa7996d5e7879e2b1290701 +size 1760547 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample167-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample167-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..635340368c1faeb3b688167ffb8ea775e8f45878 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample167-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dae6318613734bb54d2a19a011864aec567ddc08968b0871b652c75cebde9965 +size 1679208 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample173-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample173-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..8837db8c37391212a7220c7309e395b8ed7b2e69 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample173-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02ac40f02c3bc01d4ff46ceb21cbe639b28f60d5ace1be7a60e40cc7782e7d57 +size 1723088 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample175-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample175-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..2b4e32eaf9d4b5c6519ec1cd818a1dbe7e78fab2 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample175-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c55d4d5171bc2f8dc00b03c20316c9be55d9089b84c537ca374ab6b8fda3a2b7 +size 1851732 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample179-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample179-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..eb70ee977d13eb7e3b01be8859ac2c45a609029d --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample179-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:071681511750faaca7a7d68ffe366197f1cf330e4b4b29358aa58749c7f4c87f +size 1694041 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample186-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample186-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..1d5e22e3c89ef16b4bf08883b02fcd6529606446 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample186-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9c89128b4af667bccbfa4df20c5c59987b60458d73fc31abc2e95c1117ebfa1 +size 1687766 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample19-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample19-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..a370871d8aae6d6468ecc74f211cc2f3a0f8d669 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample19-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00c402394a7dc94aa6d8761fd410dffdd8f5d78595fb2b1079e3c436a1a82943 +size 1778963 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample190-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample190-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..98073e200cf3854db90b6290668deea54cd53849 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample190-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62b265816fca491f6a66738ad065672f0ab14b4cbd09f9d63843f37345dbe142 +size 1620827 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample2-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample2-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..c4e161b8243de703d985a0c029e6a75099c6a9bd --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample2-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40e93b6fd495c32dc5ce578b233e4c5295329404a22fc3018667fa9af8695d42 +size 2306604 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample20-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample20-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..67d36d37fc319e552abdf5a159a3de3f2a3a3cc2 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample20-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4dd202b61228cf34e07fa7f17c82a9f7a39ea94f338f329060c86daec0a1b8f9 +size 1787375 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample204-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample204-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..58d5c0dfc90614c12de6f2c195f11a752e7b0cd9 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample204-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d1cb0514ba25774dcd257a1940b785c496be77c064fb14d0cd18034da3fbaa55 +size 1669893 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample21-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample21-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..f6b8ae7949588ecf8a094af7de40d6dbdcb0b171 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample21-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4125d1b35cf20af2e633595ab02128d6e651d3b138c8a86841c778747ec4242 +size 1663378 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample218-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample218-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..512cba36e333f4500533f776df4b577865aaadde --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample218-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e71f09de4947206fbd26ac8e030f94eef0493f504c350299095b22775dac24e7 +size 1621348 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample22-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample22-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..80d238e200f3676d7d19e4618cb7c445032fac89 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample22-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85a627bee3e62549306a2fffb48c80430fd9e7ffbcb691aa13cfefe0700a42a5 +size 1742506 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample23-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample23-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..3c8d4ef5d3fed1e23a3d5c3cca7d9dfd69b417cb --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample23-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f999764ff62cc56d39289e772e2b434dc8fe74a6f3ef9d7485af4232a145b76 +size 1740974 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample234-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample234-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..d0a3104400295772f452c38adb956cf9247f307d --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample234-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4dfced80a54c0e26a0cebed249392d3b87c1c6c12293f11774cfeaea59457484 +size 1641001 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample24-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample24-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..5a34deb0d2f9e78b042299cf22e10ad0c0815af9 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample24-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3cde808445f6b88416f5453e622531f4e9d65d52855e623b7cb2d3b379f27be3 +size 1774196 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample25-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample25-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..43f2d089218c4a2d53f055058ff9ce8d962a3a6d --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample25-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e888531cbfff348127426bc9c030bd1fc493194a16ced70fcebb3f973e5da5c +size 1809783 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample26-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample26-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..961b6ecb3be2b5b31f65c0019edb02b57dc86179 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample26-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21c2173a9c299c1f5a1a6bd1cc6eec0ae9358628d55549f9cee61fe036a97a72 +size 1791344 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample266-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample266-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..f5c8c1fad78ae479114ffc2081ad64b49f6979ca --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample266-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60e9d285ca6c79a6dc70a96d5face9b34c240a66010c95b4e23102d6048fcf95 +size 1616172 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample27-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample27-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..25fb7103d2dd720b0b05468c845f5cf86b262ef8 --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample27-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e7a890bc0c5cfffc54d5487243960e5914f089f871f6a130f63442b16ae6a10 +size 1782045 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample283-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample283-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..2e5817af72f02a986a4f9f873643fc7a7e17700e --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample283-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5846413b4da916ded5dcd55bd81b14b3559c9e3b7819be58b7a7a89057ebd40 +size 1589842 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample29-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample29-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..596a45b0dc6c94d177a36d2a01e89ef29e5863ff --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample29-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9788d4be16b794f14d8519d7cc40acb3b92b090e61d115def06937f37f95784a +size 1755420 diff --git a/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample3-layer4-item1.zst b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample3-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..4165fd224bedbbae2efd5f3be1fd079f9e6613ff --- /dev/null +++ b/lambda0.004/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample3-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5dddef049d134c7a51e47c801a62af604afdb4308d0c0183fa6e05deb147ebe +size 2118994 diff --git a/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/dtufc_hyperprior-featurecoding_falconmamba_individual.log b/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/dtufc_hyperprior-featurecoding_falconmamba_individual.log new file mode 100644 index 0000000000000000000000000000000000000000..499edd1b1337917d1e35a55a8d40b1e2f4b606a9 --- /dev/null +++ b/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/dtufc_hyperprior-featurecoding_falconmamba_individual.log @@ -0,0 +1,16858 @@ +Experiment: dtufc_hyperprior-featurecoding_falconmamba_individual +Log file: output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/dtufc_hyperprior-featurecoding_falconmamba_individual.log +DTUFCCodecConfig: + arch: hyperprior-featurecoding + handler: falconmamba + checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.02_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar + transform_type: kmeans + transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json + bit_depth: 8 + device: cuda:0 +Loading checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.02_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Checkpoint epoch: 598 +Loaded hyperprior-featurecoding (1-channel) on cuda:0 +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.0.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.0.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.1.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.1.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.2.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.2.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.3.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.3.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.4.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.4.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_output.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.4.output' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_output.json +Loaded per-key mappings: model=falconmamba + Keys: ['layer.0.conv_state', 'layer.0.ssm_state', 'layer.1.conv_state', 'layer.1.ssm_state', 'layer.2.conv_state', 'layer.2.ssm_state', 'layer.3.conv_state', 'layer.3.ssm_state', 'layer.4.conv_state', 'layer.4.ssm_state', 'layer.4.output'] +---------------- ----------------------------------------------------------------------------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.02_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Transform type kmeans +Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json +Input ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1 +Output output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1 +---------------- ----------------------------------------------------------------------------------------------------------------------------- +Files found: 100 +---------------------------------------------------------------------- + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample0-layer4-item1.zst (1/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample0-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 275, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.014s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 275, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 275, 4096]) -> torch.Size([1, 1, 275, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,544B, BPFP=3.8784 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,096B, BPFP=2.9355 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,048B, BPFP=10.2656 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,112B, BPFP=3.9741 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,564B, BPFP=11.6123 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,816B, BPFP=3.7119 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,208B, BPFP=11.5254 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,688B, BPFP=3.4600 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,900B, BPFP=10.4736 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 291,844B, BPFP=2.0728 +⌛️ [2/4] FRONTEND: Frontend time: 0.847s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 275, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.943s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 275, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000022 0.00007236 + layer.1.conv_state 0.00048975 0.40308005 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00012844 0.06981808 + layer.3.ssm_state 0.00000001 0.00000521 + layer.3.conv_state 0.00007646 0.06322019 + layer.4.ssm_state 0.00000001 0.00000649 + layer.4.conv_state 0.00024491 0.14918499 + layer.4.output 0.00000149 0.00029576 + ------------------------------------------------------------------------------------- + TOTAL 0.00001917 0.01392685 + (elements=1,945,600) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1945600 +Total Bytes 808444 +BPFP 3.3242 bits/point +EBPFP 5.4484 equivalent bits/point +MSE 0.013927 +---------------------- -------------------------------------------------------- +Time: 1.804s Load: 0.014s, Pack+Encode: 0.847s, Decode+Unpack: 0.943s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 275, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0139 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample0-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample0-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample1-layer4-item1.zst (2/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample1-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 282, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 282, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 282, 4096]) -> torch.Size([1, 1, 282, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,712B, BPFP=3.8887 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,456B, BPFP=2.9575 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,084B, BPFP=10.2744 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,000B, BPFP=3.9673 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,364B, BPFP=11.5635 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,824B, BPFP=3.7124 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,048B, BPFP=11.4863 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,428B, BPFP=3.4441 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,708B, BPFP=10.4268 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 297,896B, BPFP=2.0632 +⌛️ [2/4] FRONTEND: Frontend time: 0.499s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 282, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.922s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 282, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000029 0.00007310 + layer.1.conv_state 0.00048919 0.40571812 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00011763 0.07058283 + layer.3.ssm_state 0.00000001 0.00000487 + layer.3.conv_state 0.00011288 0.06261945 + layer.4.ssm_state 0.00000001 0.00000602 + layer.4.conv_state 0.00028411 0.14959790 + layer.4.output 0.00000122 0.00027265 + ------------------------------------------------------------------------------------- + TOTAL 0.00001982 0.01376873 + (elements=1,974,272) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1974272 +Total Bytes 814144 +BPFP 3.2990 bits/point +EBPFP 5.3909 equivalent bits/point +MSE 0.013769 +---------------------- -------------------------------------------------------- +Time: 1.434s Load: 0.012s, Pack+Encode: 0.499s, Decode+Unpack: 0.922s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 282, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0138 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample1-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample1-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample10-layer4-item1.zst (3/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample10-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 274, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 274, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 274, 4096]) -> torch.Size([1, 1, 274, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,568B, BPFP=3.8799 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,444B, BPFP=2.9568 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,232B, BPFP=10.3105 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,084B, BPFP=3.9724 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,568B, BPFP=11.6133 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,664B, BPFP=3.7026 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,180B, BPFP=11.5186 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,196B, BPFP=3.4299 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,884B, BPFP=10.4697 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 301,792B, BPFP=2.1512 +⌛️ [2/4] FRONTEND: Frontend time: 0.466s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 274, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.925s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 274, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007422 + layer.1.conv_state 0.00049366 0.40699252 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00012530 0.07099986 + layer.3.ssm_state 0.00000001 0.00000500 + layer.3.conv_state 0.00007253 0.06328715 + layer.4.ssm_state 0.00000001 0.00000636 + layer.4.conv_state 0.00034856 0.15648827 + layer.4.output 0.00000124 0.00031314 + ------------------------------------------------------------------------------------- + TOTAL 0.00002076 0.01417613 + (elements=1,941,504) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1941504 +Total Bytes 818236 +BPFP 3.3716 bits/point +EBPFP 5.4996 equivalent bits/point +MSE 0.014176 +---------------------- -------------------------------------------------------- +Time: 1.403s Load: 0.012s, Pack+Encode: 0.466s, Decode+Unpack: 0.925s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 274, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0142 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample10-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample10-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample101-layer4-item1.zst (4/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample101-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 165, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 165, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 165, 4096]) -> torch.Size([1, 1, 165, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,528B, BPFP=3.8774 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,296B, BPFP=2.9478 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,076B, BPFP=10.2725 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,132B, BPFP=3.9753 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,468B, BPFP=11.5889 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,872B, BPFP=3.7153 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,196B, BPFP=11.5225 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,840B, BPFP=3.4692 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,920B, BPFP=10.4785 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 196,048B, BPFP=2.3206 +⌛️ [2/4] FRONTEND: Frontend time: 0.461s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 165, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.857s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 165, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007206 + layer.1.conv_state 0.00048753 0.40241110 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00013029 0.07041654 + layer.3.ssm_state 0.00000001 0.00000499 + layer.3.conv_state 0.00011288 0.06284434 + layer.4.ssm_state 0.00000001 0.00000653 + layer.4.conv_state 0.00021616 0.15130910 + layer.4.output 0.00000217 0.00037260 + ------------------------------------------------------------------------------------- + TOTAL 0.00002496 0.01810632 + (elements=1,495,040) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1495040 +Total Bytes 713000 +BPFP 3.8153 bits/point +EBPFP 6.5815 equivalent bits/point +MSE 0.018106 +---------------------- -------------------------------------------------------- +Time: 1.328s Load: 0.009s, Pack+Encode: 0.461s, Decode+Unpack: 0.857s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 165, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0181 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample101-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample101-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample102-layer4-item1.zst (5/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample102-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 162, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 162, 4096]) -> torch.Size([1, 1, 162, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,468B, BPFP=3.8738 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,524B, BPFP=2.9617 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,228B, BPFP=10.3096 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,144B, BPFP=3.9761 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,408B, BPFP=11.5742 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,732B, BPFP=3.7068 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,136B, BPFP=11.5078 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,896B, BPFP=3.4727 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,896B, BPFP=10.4727 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 191,200B, BPFP=2.3052 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 162, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.871s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 162, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007371 + layer.1.conv_state 0.00049337 0.40357882 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00015958 0.07071851 + layer.3.ssm_state 0.00000001 0.00000486 + layer.3.conv_state 0.00007099 0.06292198 + layer.4.ssm_state 0.00000001 0.00000605 + layer.4.conv_state 0.00026744 0.15170069 + layer.4.output 0.00000226 0.00043076 + ------------------------------------------------------------------------------------- + TOTAL 0.00002617 0.01832225 + (elements=1,482,752) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1482752 +Total Bytes 708256 +BPFP 3.8213 bits/point +EBPFP 6.6110 equivalent bits/point +MSE 0.018322 +---------------------- -------------------------------------------------------- +Time: 1.316s Load: 0.011s, Pack+Encode: 0.434s, Decode+Unpack: 0.871s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0183 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample102-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample102-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample106-layer4-item1.zst (6/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample106-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 185, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.013s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 185, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 185, 4096]) -> torch.Size([1, 1, 185, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,396B, BPFP=3.8694 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,420B, BPFP=2.9553 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,080B, BPFP=10.2734 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,920B, BPFP=3.9624 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,652B, BPFP=11.6338 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,632B, BPFP=3.7007 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,264B, BPFP=11.5391 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,668B, BPFP=3.4587 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,864B, BPFP=10.4648 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 207,936B, BPFP=2.1953 +⌛️ [2/4] FRONTEND: Frontend time: 0.453s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 185, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.865s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 185, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000038 0.00007198 + layer.1.conv_state 0.00049648 0.40412807 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00012672 0.07150315 + layer.3.ssm_state 0.00000001 0.00000476 + layer.3.conv_state 0.00007687 0.06374596 + layer.4.ssm_state 0.00000001 0.00000623 + layer.4.conv_state 0.00032793 0.15683775 + layer.4.output 0.00000201 0.00034385 + ------------------------------------------------------------------------------------- + TOTAL 0.00002540 0.01736309 + (elements=1,576,960) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1576960 +Total Bytes 724456 +BPFP 3.6752 bits/point +EBPFP 6.2955 equivalent bits/point +MSE 0.017363 +---------------------- -------------------------------------------------------- +Time: 1.331s Load: 0.013s, Pack+Encode: 0.453s, Decode+Unpack: 0.865s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 185, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0174 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample106-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample106-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample108-layer4-item1.zst (7/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample108-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 161, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 161, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 161, 4096]) -> torch.Size([1, 1, 161, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,404B, BPFP=3.8699 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,256B, BPFP=2.9453 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,032B, BPFP=3.9692 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,460B, BPFP=11.5869 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,676B, BPFP=3.7034 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,192B, BPFP=11.5215 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,268B, BPFP=3.4954 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,760B, BPFP=10.4395 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 185,036B, BPFP=2.2447 +⌛️ [2/4] FRONTEND: Frontend time: 0.443s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 161, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.848s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 161, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000024 0.00007235 + layer.1.conv_state 0.00050173 0.40363374 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00011340 0.07030392 + layer.3.ssm_state 0.00000001 0.00000476 + layer.3.conv_state 0.00011848 0.06261770 + layer.4.ssm_state 0.00000005 0.00000590 + layer.4.conv_state 0.00022485 0.13788289 + layer.4.output 0.00000221 0.00037017 + ------------------------------------------------------------------------------------- + TOTAL 0.00002549 0.01802372 + (elements=1,478,656) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1478656 +Total Bytes 701916 +BPFP 3.7976 bits/point +EBPFP 6.5941 equivalent bits/point +MSE 0.018024 +---------------------- -------------------------------------------------------- +Time: 1.303s Load: 0.012s, Pack+Encode: 0.443s, Decode+Unpack: 0.848s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 161, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0180 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample108-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample108-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample110-layer4-item1.zst (8/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample110-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 198, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 198, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 198, 4096]) -> torch.Size([1, 1, 198, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,480B, BPFP=3.8745 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,596B, BPFP=2.9661 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,136B, BPFP=10.2871 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,952B, BPFP=3.9644 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,648B, BPFP=11.6328 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,764B, BPFP=3.7087 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,328B, BPFP=11.5547 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,552B, BPFP=3.4517 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,056B, BPFP=10.5117 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 230,144B, BPFP=2.2702 +⌛️ [2/4] FRONTEND: Frontend time: 0.472s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 198, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.883s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 198, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007230 + layer.1.conv_state 0.00048181 0.40664598 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00010981 0.07168945 + layer.3.ssm_state 0.00000001 0.00000505 + layer.3.conv_state 0.00007216 0.06456796 + layer.4.ssm_state 0.00000002 0.00000645 + layer.4.conv_state 0.00032426 0.16031794 + layer.4.output 0.00000183 0.00041805 + ------------------------------------------------------------------------------------- + TOTAL 0.00002372 0.01698500 + (elements=1,630,208) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1630208 +Total Bytes 747280 +BPFP 3.6672 bits/point +EBPFP 6.2049 equivalent bits/point +MSE 0.016985 +---------------------- -------------------------------------------------------- +Time: 1.368s Load: 0.012s, Pack+Encode: 0.472s, Decode+Unpack: 0.883s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 198, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0170 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample110-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample110-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample115-layer4-item1.zst (9/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample115-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 170, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 170, 4096]) -> torch.Size([1, 1, 170, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,612B, BPFP=3.8826 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,508B, BPFP=2.9607 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,060B, BPFP=10.2686 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,932B, BPFP=3.9631 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,388B, BPFP=11.5693 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,352B, BPFP=3.6836 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,216B, BPFP=11.5273 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,960B, BPFP=3.4766 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,844B, BPFP=10.4600 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 195,060B, BPFP=2.2410 +⌛️ [2/4] FRONTEND: Frontend time: 0.437s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 170, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.855s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 170, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007294 + layer.1.conv_state 0.00049252 0.40384626 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00011876 0.07016941 + layer.3.ssm_state 0.00000001 0.00000496 + layer.3.conv_state 0.00011952 0.06215194 + layer.4.ssm_state 0.00000001 0.00000644 + layer.4.conv_state 0.00023872 0.15037084 + layer.4.output 0.00000204 0.00039936 + ------------------------------------------------------------------------------------- + TOTAL 0.00002508 0.01786947 + (elements=1,515,520) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1515520 +Total Bytes 711556 +BPFP 3.7561 bits/point +EBPFP 6.4825 equivalent bits/point +MSE 0.017869 +---------------------- -------------------------------------------------------- +Time: 1.300s Load: 0.009s, Pack+Encode: 0.437s, Decode+Unpack: 0.855s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0179 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample115-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample115-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample116-layer4-item1.zst (10/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample116-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 190, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 190, 4096]) -> torch.Size([1, 1, 190, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,544B, BPFP=3.8784 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,220B, BPFP=2.9431 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,172B, BPFP=10.2959 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,108B, BPFP=3.9739 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,480B, BPFP=11.5918 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,976B, BPFP=3.7217 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,240B, BPFP=11.5332 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,280B, BPFP=3.4961 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,884B, BPFP=10.4697 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 197,036B, BPFP=2.0255 +⌛️ [2/4] FRONTEND: Frontend time: 0.440s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 190, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.863s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 190, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000023 0.00007347 + layer.1.conv_state 0.00050470 0.40138823 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013213 0.06983890 + layer.3.ssm_state 0.00000001 0.00000493 + layer.3.conv_state 0.00011581 0.06209447 + layer.4.ssm_state 0.00000003 0.00000585 + layer.4.conv_state 0.00021685 0.14446238 + layer.4.output 0.00000199 0.00033967 + ------------------------------------------------------------------------------------- + TOTAL 0.00002388 0.01676489 + (elements=1,597,440) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1597440 +Total Bytes 714564 +BPFP 3.5785 bits/point +EBPFP 6.1703 equivalent bits/point +MSE 0.016765 +---------------------- -------------------------------------------------------- +Time: 1.313s Load: 0.010s, Pack+Encode: 0.440s, Decode+Unpack: 0.863s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0168 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample116-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample116-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample12-layer4-item1.zst (11/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample12-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 231, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 231, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 231, 4096]) -> torch.Size([1, 1, 231, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,748B, BPFP=3.8909 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,224B, BPFP=2.9434 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,064B, BPFP=10.2695 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,992B, BPFP=3.9668 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,492B, BPFP=11.5947 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,928B, BPFP=3.7188 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,348B, BPFP=11.5596 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,036B, BPFP=3.4812 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,988B, BPFP=10.4951 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 253,996B, BPFP=2.1476 +⌛️ [2/4] FRONTEND: Frontend time: 0.455s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 231, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.884s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 231, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000024 0.00007282 + layer.1.conv_state 0.00050504 0.40419310 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00012981 0.07036899 + layer.3.ssm_state 0.00000001 0.00000495 + layer.3.conv_state 0.00007554 0.06284906 + layer.4.ssm_state 0.00000001 0.00000647 + layer.4.conv_state 0.00028923 0.15493335 + layer.4.output 0.00000157 0.00033872 + ------------------------------------------------------------------------------------- + TOTAL 0.00002213 0.01547216 + (elements=1,765,376) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1765376 +Total Bytes 771440 +BPFP 3.4959 bits/point +EBPFP 5.8407 equivalent bits/point +MSE 0.015472 +---------------------- -------------------------------------------------------- +Time: 1.350s Load: 0.011s, Pack+Encode: 0.455s, Decode+Unpack: 0.884s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 231, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0155 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample12-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample12-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample120-layer4-item1.zst (12/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample120-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 183, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 183, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) -> torch.Size([1, 1, 183, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,616B, BPFP=3.8828 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,080B, BPFP=2.9346 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,092B, BPFP=10.2764 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,896B, BPFP=3.9609 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,408B, BPFP=11.5742 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,632B, BPFP=3.7007 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,332B, BPFP=11.5557 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,968B, BPFP=3.4771 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,120B, BPFP=10.5273 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 201,192B, BPFP=2.1473 +⌛️ [2/4] FRONTEND: Frontend time: 0.432s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.850s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000024 0.00007356 + layer.1.conv_state 0.00048839 0.40368360 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00008466 0.07010557 + layer.3.ssm_state 0.00000001 0.00000495 + layer.3.conv_state 0.00011958 0.06285749 + layer.4.ssm_state 0.00000001 0.00000651 + layer.4.conv_state 0.00031804 0.15429552 + layer.4.output 0.00000201 0.00040714 + ------------------------------------------------------------------------------------- + TOTAL 0.00002515 0.01737224 + (elements=1,568,768) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1568768 +Total Bytes 717960 +BPFP 3.6613 bits/point +EBPFP 6.2965 equivalent bits/point +MSE 0.017372 +---------------------- -------------------------------------------------------- +Time: 1.291s Load: 0.009s, Pack+Encode: 0.432s, Decode+Unpack: 0.850s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 183, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0174 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample120-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample120-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample122-layer4-item1.zst (13/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample122-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 191, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 191, 4096]) -> torch.Size([1, 1, 191, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,512B, BPFP=3.8765 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,596B, BPFP=2.9661 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,180B, BPFP=10.2979 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,064B, BPFP=3.9712 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,468B, BPFP=11.5889 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,064B, BPFP=3.7271 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,124B, BPFP=11.5049 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 58,028B, BPFP=3.5417 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,768B, BPFP=10.4414 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 205,756B, BPFP=2.1040 +⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 191, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.872s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 191, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007281 + layer.1.conv_state 0.00046948 0.40452367 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00011482 0.07068680 + layer.3.ssm_state 0.00000001 0.00000532 + layer.3.conv_state 0.00011212 0.06242863 + layer.4.ssm_state 0.00000002 0.00000561 + layer.4.conv_state 0.00024277 0.13408954 + layer.4.output 0.00000205 0.00033806 + ------------------------------------------------------------------------------------- + TOTAL 0.00002323 0.01659816 + (elements=1,601,536) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1601536 +Total Bytes 724184 +BPFP 3.6174 bits/point +EBPFP 6.2071 equivalent bits/point +MSE 0.016598 +---------------------- -------------------------------------------------------- +Time: 1.316s Load: 0.009s, Pack+Encode: 0.435s, Decode+Unpack: 0.872s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0166 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample122-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample122-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample127-layer4-item1.zst (14/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample127-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 175, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 175, 4096]) -> torch.Size([1, 1, 175, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,760B, BPFP=3.8916 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,424B, BPFP=2.9556 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,160B, BPFP=10.2930 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,092B, BPFP=3.9729 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,496B, BPFP=11.5957 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,944B, BPFP=3.7197 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,200B, BPFP=11.5234 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,508B, BPFP=3.4490 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,780B, BPFP=10.4443 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 203,384B, BPFP=2.2699 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 175, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.931s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 175, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007278 + layer.1.conv_state 0.00051384 0.40145260 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00012234 0.07004829 + layer.3.ssm_state 0.00000001 0.00000505 + layer.3.conv_state 0.00007815 0.06273685 + layer.4.ssm_state 0.00000001 0.00000643 + layer.4.conv_state 0.00034266 0.15818961 + layer.4.output 0.00000210 0.00039944 + ------------------------------------------------------------------------------------- + TOTAL 0.00002667 0.01776220 + (elements=1,536,000) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1536000 +Total Bytes 720372 +BPFP 3.7519 bits/point +EBPFP 6.4446 equivalent bits/point +MSE 0.017762 +---------------------- -------------------------------------------------------- +Time: 1.377s Load: 0.012s, Pack+Encode: 0.434s, Decode+Unpack: 0.931s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0178 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample127-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample127-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample128-layer4-item1.zst (15/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample128-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 168, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) -> torch.Size([1, 1, 168, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,652B, BPFP=3.8850 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,436B, BPFP=2.9563 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,100B, BPFP=10.2783 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,048B, BPFP=3.9702 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,540B, BPFP=11.6064 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,784B, BPFP=3.7100 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,144B, BPFP=11.5098 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,240B, BPFP=3.4937 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,752B, BPFP=10.4375 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 196,144B, BPFP=2.2803 +⌛️ [2/4] FRONTEND: Frontend time: 0.530s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.859s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007158 + layer.1.conv_state 0.00051486 0.40193421 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00016795 0.06979756 + layer.3.ssm_state 0.00000001 0.00000496 + layer.3.conv_state 0.00011430 0.06259174 + layer.4.ssm_state 0.00000002 0.00000602 + layer.4.conv_state 0.00021850 0.14201021 + layer.4.output 0.00000215 0.00034889 + ------------------------------------------------------------------------------------- + TOTAL 0.00002626 0.01771937 + (elements=1,507,328) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1507328 +Total Bytes 713464 +BPFP 3.7866 bits/point +EBPFP 6.5323 equivalent bits/point +MSE 0.017719 +---------------------- -------------------------------------------------------- +Time: 1.400s Load: 0.011s, Pack+Encode: 0.530s, Decode+Unpack: 0.859s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0177 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample128-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample128-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample129-layer4-item1.zst (16/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample129-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 181, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 181, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 181, 4096]) -> torch.Size([1, 1, 181, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,624B, BPFP=3.8833 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,240B, BPFP=2.9443 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,080B, BPFP=10.2734 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,020B, BPFP=3.9685 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,396B, BPFP=11.5713 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,684B, BPFP=3.7039 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,144B, BPFP=11.5098 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,712B, BPFP=3.4614 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,928B, BPFP=10.4805 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 200,764B, BPFP=2.1664 +⌛️ [2/4] FRONTEND: Frontend time: 0.436s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 181, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.849s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 181, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007234 + layer.1.conv_state 0.00048726 0.40487665 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00012801 0.06953393 + layer.3.ssm_state 0.00000001 0.00000495 + layer.3.conv_state 0.00011944 0.06193350 + layer.4.ssm_state 0.00000001 0.00000635 + layer.4.conv_state 0.00034131 0.15545076 + layer.4.output 0.00000206 0.00032365 + ------------------------------------------------------------------------------------- + TOTAL 0.00002667 0.01743943 + (elements=1,560,576) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1560576 +Total Bytes 717216 +BPFP 3.6767 bits/point +EBPFP 6.3242 equivalent bits/point +MSE 0.017439 +---------------------- -------------------------------------------------------- +Time: 1.298s Load: 0.012s, Pack+Encode: 0.436s, Decode+Unpack: 0.849s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 181, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0174 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample129-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample129-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample130-layer4-item1.zst (17/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample130-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 166, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) -> torch.Size([1, 1, 166, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,624B, BPFP=3.8833 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,368B, BPFP=2.9521 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,116B, BPFP=10.2822 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,060B, BPFP=3.9709 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,524B, BPFP=11.6025 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,000B, BPFP=3.7231 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,216B, BPFP=11.5273 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,552B, BPFP=3.4517 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,804B, BPFP=10.4502 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 194,952B, BPFP=2.2938 +⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.861s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007062 + layer.1.conv_state 0.00049597 0.40483737 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00012619 0.07094771 + layer.3.ssm_state 0.00000001 0.00000522 + layer.3.conv_state 0.00006821 0.06264811 + layer.4.ssm_state 0.00000001 0.00000660 + layer.4.conv_state 0.00029570 0.15519427 + layer.4.output 0.00000206 0.00042929 + ------------------------------------------------------------------------------------- + TOTAL 0.00002569 0.01822876 + (elements=1,499,136) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1499136 +Total Bytes 711840 +BPFP 3.7987 bits/point +EBPFP 6.5570 equivalent bits/point +MSE 0.018229 +---------------------- -------------------------------------------------------- +Time: 1.308s Load: 0.012s, Pack+Encode: 0.435s, Decode+Unpack: 0.861s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0182 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample130-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample130-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample134-layer4-item1.zst (18/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample134-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 182, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 182, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 182, 4096]) -> torch.Size([1, 1, 182, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,576B, BPFP=3.8804 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,332B, BPFP=2.9500 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,232B, BPFP=10.3105 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,036B, BPFP=3.9695 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,392B, BPFP=11.5703 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,588B, BPFP=3.6980 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,160B, BPFP=11.5137 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,000B, BPFP=3.4790 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,992B, BPFP=10.4961 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 206,268B, BPFP=2.2136 +⌛️ [2/4] FRONTEND: Frontend time: 0.440s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 182, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.861s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 182, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007307 + layer.1.conv_state 0.00049604 0.40347368 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00012832 0.06988574 + layer.3.ssm_state 0.00000001 0.00000476 + layer.3.conv_state 0.00011179 0.06264901 + layer.4.ssm_state 0.00000001 0.00000625 + layer.4.conv_state 0.00022786 0.14441817 + layer.4.output 0.00000216 0.00043120 + ------------------------------------------------------------------------------------- + TOTAL 0.00002431 0.01720782 + (elements=1,564,672) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1564672 +Total Bytes 723200 +BPFP 3.6976 bits/point +EBPFP 6.3407 equivalent bits/point +MSE 0.017208 +---------------------- -------------------------------------------------------- +Time: 1.313s Load: 0.012s, Pack+Encode: 0.440s, Decode+Unpack: 0.861s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 182, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0172 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample134-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample134-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample135-layer4-item1.zst (19/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample135-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 178, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 178, 4096]) -> torch.Size([1, 1, 178, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,628B, BPFP=3.8835 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,348B, BPFP=2.9509 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,148B, BPFP=10.2900 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,040B, BPFP=3.9697 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,472B, BPFP=11.5898 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,604B, BPFP=3.6990 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,212B, BPFP=11.5264 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,280B, BPFP=3.4351 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,008B, BPFP=10.5000 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 205,804B, BPFP=2.2582 +⌛️ [2/4] FRONTEND: Frontend time: 0.444s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 178, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.854s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 178, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007246 + layer.1.conv_state 0.00048610 0.40686682 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00014962 0.07073726 + layer.3.ssm_state 0.00000001 0.00000502 + layer.3.conv_state 0.00007021 0.06304121 + layer.4.ssm_state 0.00000001 0.00000635 + layer.4.conv_state 0.00031005 0.15346932 + layer.4.output 0.00000198 0.00042373 + ------------------------------------------------------------------------------------- + TOTAL 0.00002555 0.01767151 + (elements=1,548,288) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1548288 +Total Bytes 722168 +BPFP 3.7314 bits/point +EBPFP 6.3995 equivalent bits/point +MSE 0.017672 +---------------------- -------------------------------------------------------- +Time: 1.309s Load: 0.011s, Pack+Encode: 0.444s, Decode+Unpack: 0.854s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0177 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample135-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample135-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample142-layer4-item1.zst (20/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample142-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 166, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) -> torch.Size([1, 1, 166, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,596B, BPFP=3.8816 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,176B, BPFP=2.9404 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,180B, BPFP=10.2979 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,884B, BPFP=3.9602 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,516B, BPFP=11.6006 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,808B, BPFP=3.7114 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,168B, BPFP=11.5156 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,280B, BPFP=3.4961 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,868B, BPFP=10.4658 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 193,492B, BPFP=2.2766 +⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.857s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007207 + layer.1.conv_state 0.00050515 0.40283424 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013851 0.07096175 + layer.3.ssm_state 0.00000001 0.00000497 + layer.3.conv_state 0.00011516 0.06293257 + layer.4.ssm_state 0.00000001 0.00000575 + layer.4.conv_state 0.00022993 0.13894282 + layer.4.output 0.00000208 0.00044904 + ------------------------------------------------------------------------------------- + TOTAL 0.00002578 0.01784526 + (elements=1,499,136) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1499136 +Total Bytes 710592 +BPFP 3.7920 bits/point +EBPFP 6.5515 equivalent bits/point +MSE 0.017845 +---------------------- -------------------------------------------------------- +Time: 1.302s Load: 0.011s, Pack+Encode: 0.435s, Decode+Unpack: 0.857s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0178 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample142-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample142-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample143-layer4-item1.zst (21/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample143-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 154, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 154, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 154, 4096]) -> torch.Size([1, 1, 154, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,476B, BPFP=3.8743 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,360B, BPFP=2.9517 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,128B, BPFP=10.2852 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,780B, BPFP=3.9539 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,396B, BPFP=11.5713 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,520B, BPFP=3.6938 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,112B, BPFP=11.5020 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,936B, BPFP=3.4751 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,052B, BPFP=10.5107 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 182,576B, BPFP=2.3155 +⌛️ [2/4] FRONTEND: Frontend time: 0.436s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 154, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.859s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 154, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007221 + layer.1.conv_state 0.00051446 0.40034059 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00012903 0.07069310 + layer.3.ssm_state 0.00000001 0.00000498 + layer.3.conv_state 0.00007035 0.06272963 + layer.4.ssm_state 0.00000004 0.00000613 + layer.4.conv_state 0.00022206 0.14456974 + layer.4.output 0.00000218 0.00044051 + ------------------------------------------------------------------------------------- + TOTAL 0.00002543 0.01849145 + (elements=1,449,984) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1449984 +Total Bytes 698960 +BPFP 3.8564 bits/point +EBPFP 6.7054 equivalent bits/point +MSE 0.018491 +---------------------- -------------------------------------------------------- +Time: 1.306s Load: 0.011s, Pack+Encode: 0.436s, Decode+Unpack: 0.859s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 154, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0185 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample143-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample143-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample144-layer4-item1.zst (22/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample144-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 166, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) -> torch.Size([1, 1, 166, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,628B, BPFP=3.8835 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,432B, BPFP=2.9561 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,100B, BPFP=10.2783 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,112B, BPFP=3.9741 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,480B, BPFP=11.5918 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,692B, BPFP=3.7043 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,052B, BPFP=11.4873 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,616B, BPFP=3.4556 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,940B, BPFP=10.4834 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 194,368B, BPFP=2.2869 +⌛️ [2/4] FRONTEND: Frontend time: 0.436s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.857s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007274 + layer.1.conv_state 0.00048448 0.40224779 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00013520 0.06989998 + layer.3.ssm_state 0.00000001 0.00000500 + layer.3.conv_state 0.00007505 0.06275491 + layer.4.ssm_state 0.00000002 0.00000658 + layer.4.conv_state 0.00034342 0.15796761 + layer.4.output 0.00000207 0.00043242 + ------------------------------------------------------------------------------------- + TOTAL 0.00002685 0.01821379 + (elements=1,499,136) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1499136 +Total Bytes 711044 +BPFP 3.7944 bits/point +EBPFP 6.5516 equivalent bits/point +MSE 0.018214 +---------------------- -------------------------------------------------------- +Time: 1.302s Load: 0.009s, Pack+Encode: 0.436s, Decode+Unpack: 0.857s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0182 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample144-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample144-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample148-layer4-item1.zst (23/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample148-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 197, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 197, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 197, 4096]) -> torch.Size([1, 1, 197, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,552B, BPFP=3.8789 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,396B, BPFP=2.9539 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,128B, BPFP=10.2852 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,764B, BPFP=3.9529 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,612B, BPFP=11.6240 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,500B, BPFP=3.6926 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,132B, BPFP=11.5068 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,076B, BPFP=3.4836 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,856B, BPFP=10.4629 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 234,900B, BPFP=2.3289 +⌛️ [2/4] FRONTEND: Frontend time: 0.456s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 197, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.885s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 197, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007130 + layer.1.conv_state 0.00049519 0.40445304 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00014288 0.07144357 + layer.3.ssm_state 0.00000001 0.00000500 + layer.3.conv_state 0.00007032 0.06353003 + layer.4.ssm_state 0.00000002 0.00000603 + layer.4.conv_state 0.00023237 0.14798824 + layer.4.output 0.00000183 0.00033162 + ------------------------------------------------------------------------------------- + TOTAL 0.00002283 0.01666520 + (elements=1,626,112) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1626112 +Total Bytes 751540 +BPFP 3.6974 bits/point +EBPFP 6.2391 equivalent bits/point +MSE 0.016665 +---------------------- -------------------------------------------------------- +Time: 1.354s Load: 0.012s, Pack+Encode: 0.456s, Decode+Unpack: 0.885s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 197, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0167 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample148-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample148-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample149-layer4-item1.zst (24/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample149-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 168, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) -> torch.Size([1, 1, 168, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,608B, BPFP=3.8823 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,412B, BPFP=2.9548 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,080B, BPFP=10.2734 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,948B, BPFP=3.9641 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,464B, BPFP=11.5879 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,520B, BPFP=3.6938 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,192B, BPFP=11.5215 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,116B, BPFP=3.4861 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,840B, BPFP=10.4590 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 195,308B, BPFP=2.2706 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.851s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007328 + layer.1.conv_state 0.00048592 0.40274999 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00011654 0.07057983 + layer.3.ssm_state 0.00000001 0.00000490 + layer.3.conv_state 0.00011837 0.06230515 + layer.4.ssm_state 0.00000001 0.00000620 + layer.4.conv_state 0.00020779 0.14568546 + layer.4.output 0.00000221 0.00034468 + ------------------------------------------------------------------------------------- + TOTAL 0.00002440 0.01782602 + (elements=1,507,328) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1507328 +Total Bytes 712112 +BPFP 3.7795 bits/point +EBPFP 6.5224 equivalent bits/point +MSE 0.017826 +---------------------- -------------------------------------------------------- +Time: 1.297s Load: 0.011s, Pack+Encode: 0.434s, Decode+Unpack: 0.851s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0178 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample149-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample149-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample153-layer4-item1.zst (25/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample153-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 164, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 164, 4096]) -> torch.Size([1, 1, 164, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,564B, BPFP=3.8796 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,060B, BPFP=2.9333 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,140B, BPFP=10.2881 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,920B, BPFP=3.9624 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,628B, BPFP=11.6279 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,928B, BPFP=3.7188 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,176B, BPFP=11.5176 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,240B, BPFP=3.4937 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,928B, BPFP=10.4805 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 191,040B, BPFP=2.2752 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 164, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.859s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 164, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000029 0.00007266 + layer.1.conv_state 0.00049823 0.40509194 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00014568 0.07101848 + layer.3.ssm_state 0.00000001 0.00000505 + layer.3.conv_state 0.00011339 0.06316011 + layer.4.ssm_state 0.00000002 0.00000599 + layer.4.conv_state 0.00023177 0.14136972 + layer.4.output 0.00000211 0.00035519 + ------------------------------------------------------------------------------------- + TOTAL 0.00002593 0.01800785 + (elements=1,490,944) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1490944 +Total Bytes 708248 +BPFP 3.8003 bits/point +EBPFP 6.5755 equivalent bits/point +MSE 0.018008 +---------------------- -------------------------------------------------------- +Time: 1.305s Load: 0.012s, Pack+Encode: 0.434s, Decode+Unpack: 0.859s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0180 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample153-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample153-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample156-layer4-item1.zst (26/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample156-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 163, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 163, 4096]) -> torch.Size([1, 1, 163, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,604B, BPFP=3.8821 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,516B, BPFP=2.9612 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,048B, BPFP=10.2656 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,044B, BPFP=3.9700 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,492B, BPFP=11.5947 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,916B, BPFP=3.7180 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,160B, BPFP=11.5137 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,984B, BPFP=3.4780 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,812B, BPFP=10.4521 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 189,124B, BPFP=2.2662 +⌛️ [2/4] FRONTEND: Frontend time: 0.436s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 163, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.882s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 163, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000043 0.00007150 + layer.1.conv_state 0.00049103 0.40312722 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00014094 0.07054198 + layer.3.ssm_state 0.00000001 0.00000518 + layer.3.conv_state 0.00012075 0.06271178 + layer.4.ssm_state 0.00000001 0.00000631 + layer.4.conv_state 0.00020319 0.15074420 + layer.4.output 0.00000221 0.00044347 + ------------------------------------------------------------------------------------- + TOTAL 0.00002532 0.01823898 + (elements=1,486,848) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1486848 +Total Bytes 706324 +BPFP 3.8004 bits/point +EBPFP 6.5832 equivalent bits/point +MSE 0.018239 +---------------------- -------------------------------------------------------- +Time: 1.328s Load: 0.010s, Pack+Encode: 0.436s, Decode+Unpack: 0.882s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 163, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0182 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample156-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample156-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample158-layer4-item1.zst (27/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample158-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 162, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 162, 4096]) -> torch.Size([1, 1, 162, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,776B, BPFP=3.8926 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,452B, BPFP=2.9573 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,120B, BPFP=10.2832 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,200B, BPFP=3.9795 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,460B, BPFP=11.5869 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,272B, BPFP=3.7397 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,260B, BPFP=11.5381 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,184B, BPFP=3.4902 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,072B, BPFP=10.5156 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 186,620B, BPFP=2.2500 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 162, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.854s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 162, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000256 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007393 + layer.1.conv_state 0.00050753 0.40154514 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00012648 0.07011558 + layer.3.ssm_state 0.00000001 0.00000521 + layer.3.conv_state 0.00011960 0.06363945 + layer.4.ssm_state 0.00000001 0.00000644 + layer.4.conv_state 0.00022111 0.14860524 + layer.4.output 0.00000221 0.00041997 + ------------------------------------------------------------------------------------- + TOTAL 0.00002578 0.01820670 + (elements=1,482,752) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1482752 +Total Bytes 705040 +BPFP 3.8040 bits/point +EBPFP 6.6010 equivalent bits/point +MSE 0.018207 +---------------------- -------------------------------------------------------- +Time: 1.299s Load: 0.011s, Pack+Encode: 0.434s, Decode+Unpack: 0.854s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0182 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample158-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample158-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample159-layer4-item1.zst (28/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample159-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 162, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 162, 4096]) -> torch.Size([1, 1, 162, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,580B, BPFP=3.8806 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,304B, BPFP=2.9482 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,116B, BPFP=10.2822 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,168B, BPFP=3.9775 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,532B, BPFP=11.6045 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,800B, BPFP=3.7109 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,140B, BPFP=11.5088 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,012B, BPFP=3.4797 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,700B, BPFP=10.4248 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 186,124B, BPFP=2.2440 +⌛️ [2/4] FRONTEND: Frontend time: 0.432s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 162, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.869s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 162, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007316 + layer.1.conv_state 0.00049392 0.40234974 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00013642 0.06970318 + layer.3.ssm_state 0.00000001 0.00000480 + layer.3.conv_state 0.00011766 0.06171326 + layer.4.ssm_state 0.00000003 0.00000594 + layer.4.conv_state 0.00021047 0.14626917 + layer.4.output 0.00000223 0.00042807 + ------------------------------------------------------------------------------------- + TOTAL 0.00002544 0.01812464 + (elements=1,482,752) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1482752 +Total Bytes 703100 +BPFP 3.7935 bits/point +EBPFP 6.5828 equivalent bits/point +MSE 0.018125 +---------------------- -------------------------------------------------------- +Time: 1.313s Load: 0.012s, Pack+Encode: 0.432s, Decode+Unpack: 0.869s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 162, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0181 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample159-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample159-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample162-layer4-item1.zst (29/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample162-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 151, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 151, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 151, 4096]) -> torch.Size([1, 1, 151, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,560B, BPFP=3.8794 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,360B, BPFP=2.9517 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,128B, BPFP=10.2852 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,068B, BPFP=3.9714 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,500B, BPFP=11.5967 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,732B, BPFP=3.7068 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,216B, BPFP=11.5273 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,820B, BPFP=3.4680 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,000B, BPFP=10.4980 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 179,972B, BPFP=2.3279 +⌛️ [2/4] FRONTEND: Frontend time: 0.449s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 151, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.853s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 151, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007214 + layer.1.conv_state 0.00048080 0.40184548 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00015695 0.06996829 + layer.3.ssm_state 0.00000001 0.00000508 + layer.3.conv_state 0.00007366 0.06333081 + layer.4.ssm_state 0.00000001 0.00000637 + layer.4.conv_state 0.00025051 0.15297391 + layer.4.output 0.00000211 0.00050472 + ------------------------------------------------------------------------------------- + TOTAL 0.00002617 0.01889641 + (elements=1,437,696) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1437696 +Total Bytes 696980 +BPFP 3.8783 bits/point +EBPFP 6.7552 equivalent bits/point +MSE 0.018896 +---------------------- -------------------------------------------------------- +Time: 1.314s Load: 0.011s, Pack+Encode: 0.449s, Decode+Unpack: 0.853s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 151, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0189 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample162-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample162-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample166-layer4-item1.zst (30/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample166-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 188, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 188, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 188, 4096]) -> torch.Size([1, 1, 188, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,640B, BPFP=3.8843 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,148B, BPFP=2.9387 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,236B, BPFP=10.3115 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,044B, BPFP=3.9700 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,588B, BPFP=11.6182 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,836B, BPFP=3.7131 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,248B, BPFP=11.5352 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,076B, BPFP=3.4226 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,100B, BPFP=10.5225 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 205,456B, BPFP=2.1345 +⌛️ [2/4] FRONTEND: Frontend time: 0.433s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 188, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.848s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 188, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007048 + layer.1.conv_state 0.00047809 0.40537268 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00013752 0.07212804 + layer.3.ssm_state 0.00000001 0.00000505 + layer.3.conv_state 0.00007009 0.06480289 + layer.4.ssm_state 0.00000002 0.00000662 + layer.4.conv_state 0.00055818 0.16452746 + layer.4.output 0.00000209 0.00033462 + ------------------------------------------------------------------------------------- + TOTAL 0.00002970 0.01744585 + (elements=1,589,248) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1589248 +Total Bytes 721996 +BPFP 3.6344 bits/point +EBPFP 6.2346 equivalent bits/point +MSE 0.017446 +---------------------- -------------------------------------------------------- +Time: 1.290s Load: 0.009s, Pack+Encode: 0.433s, Decode+Unpack: 0.848s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 188, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0174 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample166-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample166-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample167-layer4-item1.zst (31/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample167-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 173, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 173, 4096]) -> torch.Size([1, 1, 173, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,684B, BPFP=3.8870 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,180B, BPFP=2.9407 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,124B, BPFP=10.2842 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,808B, BPFP=3.9556 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,536B, BPFP=11.6055 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,804B, BPFP=3.7112 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,208B, BPFP=11.5254 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,544B, BPFP=3.4512 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,980B, BPFP=10.4932 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 196,872B, BPFP=2.2226 +⌛️ [2/4] FRONTEND: Frontend time: 0.433s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 173, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.853s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 173, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000255 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007354 + layer.1.conv_state 0.00049089 0.40271264 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013154 0.06950472 + layer.3.ssm_state 0.00000001 0.00000497 + layer.3.conv_state 0.00011542 0.06201870 + layer.4.ssm_state 0.00000001 0.00000625 + layer.4.conv_state 0.00026976 0.14966837 + layer.4.output 0.00000213 0.00033098 + ------------------------------------------------------------------------------------- + TOTAL 0.00002576 0.01764078 + (elements=1,527,808) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1527808 +Total Bytes 713364 +BPFP 3.7354 bits/point +EBPFP 6.4398 equivalent bits/point +MSE 0.017641 +---------------------- -------------------------------------------------------- +Time: 1.297s Load: 0.011s, Pack+Encode: 0.433s, Decode+Unpack: 0.853s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0176 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample167-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample167-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample173-layer4-item1.zst (32/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample173-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 181, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 181, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 181, 4096]) -> torch.Size([1, 1, 181, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,632B, BPFP=3.8838 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,600B, BPFP=2.9663 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,160B, BPFP=10.2930 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,132B, BPFP=3.9753 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,536B, BPFP=11.6055 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,460B, BPFP=3.6902 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,084B, BPFP=11.4951 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 55,636B, BPFP=3.3958 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,692B, BPFP=10.4229 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 203,292B, BPFP=2.1937 +⌛️ [2/4] FRONTEND: Frontend time: 0.437s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 181, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.850s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 181, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000255 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007066 + layer.1.conv_state 0.00048935 0.40353060 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00014754 0.06996830 + layer.3.ssm_state 0.00000001 0.00000502 + layer.3.conv_state 0.00011438 0.06220390 + layer.4.ssm_state 0.00000005 0.00000675 + layer.4.conv_state 0.00050179 0.16593142 + layer.4.output 0.00000190 0.00032465 + ------------------------------------------------------------------------------------- + TOTAL 0.00003030 0.01764640 + (elements=1,560,576) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1560576 +Total Bytes 718848 +BPFP 3.6850 bits/point +EBPFP 6.3279 equivalent bits/point +MSE 0.017646 +---------------------- -------------------------------------------------------- +Time: 1.295s Load: 0.009s, Pack+Encode: 0.437s, Decode+Unpack: 0.850s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 181, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0176 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample173-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample173-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample175-layer4-item1.zst (33/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample175-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 209, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 209, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 209, 4096]) -> torch.Size([1, 1, 209, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,400B, BPFP=3.8696 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,332B, BPFP=2.9500 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,204B, BPFP=10.3037 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,940B, BPFP=3.9636 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,536B, BPFP=11.6055 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,664B, BPFP=3.7026 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,172B, BPFP=11.5166 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,064B, BPFP=3.4829 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,936B, BPFP=10.4824 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 235,104B, BPFP=2.1971 +⌛️ [2/4] FRONTEND: Frontend time: 0.460s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 209, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.894s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 209, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000014 0.00007235 + layer.1.conv_state 0.00048877 0.40529233 + layer.2.ssm_state 0.00000001 0.00000370 + layer.2.conv_state 0.00012277 0.07079140 + layer.3.ssm_state 0.00000001 0.00000493 + layer.3.conv_state 0.00012099 0.06347868 + layer.4.ssm_state 0.00000001 0.00000614 + layer.4.conv_state 0.00025427 0.14862171 + layer.4.output 0.00000161 0.00031404 + ------------------------------------------------------------------------------------- + TOTAL 0.00002299 0.01619214 + (elements=1,675,264) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1675264 +Total Bytes 751976 +BPFP 3.5910 bits/point +EBPFP 6.0592 equivalent bits/point +MSE 0.016192 +---------------------- -------------------------------------------------------- +Time: 1.364s Load: 0.010s, Pack+Encode: 0.460s, Decode+Unpack: 0.894s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 209, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0162 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample175-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample175-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample179-layer4-item1.zst (34/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample179-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 175, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 175, 4096]) -> torch.Size([1, 1, 175, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,652B, BPFP=3.8850 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,460B, BPFP=2.9578 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,108B, BPFP=10.2803 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,000B, BPFP=3.9673 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,512B, BPFP=11.5996 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,804B, BPFP=3.7112 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,052B, BPFP=11.4873 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,124B, BPFP=3.4866 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,928B, BPFP=10.4805 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 200,480B, BPFP=2.2375 +⌛️ [2/4] FRONTEND: Frontend time: 0.441s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 175, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.855s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 175, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007144 + layer.1.conv_state 0.00049563 0.40100461 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00014857 0.07070201 + layer.3.ssm_state 0.00000001 0.00000517 + layer.3.conv_state 0.00007001 0.06266624 + layer.4.ssm_state 0.00000002 0.00000600 + layer.4.conv_state 0.00020707 0.14235893 + layer.4.output 0.00000204 0.00038883 + ------------------------------------------------------------------------------------- + TOTAL 0.00002375 0.01742226 + (elements=1,536,000) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1536000 +Total Bytes 717744 +BPFP 3.7382 bits/point +EBPFP 6.4323 equivalent bits/point +MSE 0.017422 +---------------------- -------------------------------------------------------- +Time: 1.306s Load: 0.009s, Pack+Encode: 0.441s, Decode+Unpack: 0.855s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0174 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample179-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample179-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample186-layer4-item1.zst (35/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample186-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 172, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 172, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 172, 4096]) -> torch.Size([1, 1, 172, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,740B, BPFP=3.8904 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,472B, BPFP=2.9585 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,032B, BPFP=10.2617 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,992B, BPFP=3.9668 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,520B, BPFP=11.6016 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,928B, BPFP=3.7188 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,160B, BPFP=11.5137 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,404B, BPFP=3.5037 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,980B, BPFP=10.4932 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 202,172B, BPFP=2.2957 +⌛️ [2/4] FRONTEND: Frontend time: 0.440s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 172, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.858s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 172, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007246 + layer.1.conv_state 0.00050694 0.40262765 + layer.2.ssm_state 0.00000001 0.00000368 + layer.2.conv_state 0.00012346 0.07087217 + layer.3.ssm_state 0.00000001 0.00000495 + layer.3.conv_state 0.00011101 0.06250076 + layer.4.ssm_state 0.00000002 0.00000616 + layer.4.conv_state 0.00022697 0.14017259 + layer.4.output 0.00000216 0.00035163 + ------------------------------------------------------------------------------------- + TOTAL 0.00002499 0.01753050 + (elements=1,523,712) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1523712 +Total Bytes 720024 +BPFP 3.7804 bits/point +EBPFP 6.4993 equivalent bits/point +MSE 0.017530 +---------------------- -------------------------------------------------------- +Time: 1.311s Load: 0.012s, Pack+Encode: 0.440s, Decode+Unpack: 0.858s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 172, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0175 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample186-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample186-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample19-layer4-item1.zst (36/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample19-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 193, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.013s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 193, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 193, 4096]) -> torch.Size([1, 1, 193, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,568B, BPFP=3.8799 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,168B, BPFP=2.9399 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,048B, BPFP=10.2656 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,072B, BPFP=3.9717 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,452B, BPFP=11.5850 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,656B, BPFP=3.7021 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,172B, BPFP=11.5166 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,636B, BPFP=3.4568 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,784B, BPFP=10.4453 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 216,528B, BPFP=2.1912 +⌛️ [2/4] FRONTEND: Frontend time: 0.450s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 193, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.887s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 193, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000031 0.00007197 + layer.1.conv_state 0.00048110 0.40513945 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00012031 0.07058483 + layer.3.ssm_state 0.00000001 0.00000512 + layer.3.conv_state 0.00007804 0.06270272 + layer.4.ssm_state 0.00000001 0.00000639 + layer.4.conv_state 0.00027310 0.15195061 + layer.4.output 0.00000181 0.00033945 + ------------------------------------------------------------------------------------- + TOTAL 0.00002328 0.01689570 + (elements=1,609,728) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1609728 +Total Bytes 732708 +BPFP 3.6414 bits/point +EBPFP 6.2067 equivalent bits/point +MSE 0.016896 +---------------------- -------------------------------------------------------- +Time: 1.349s Load: 0.013s, Pack+Encode: 0.450s, Decode+Unpack: 0.887s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 193, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0169 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample19-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample19-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample190-layer4-item1.zst (37/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample190-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 159, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 159, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 159, 4096]) -> torch.Size([1, 1, 159, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,536B, BPFP=3.8779 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,308B, BPFP=2.9485 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,140B, BPFP=10.2881 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,008B, BPFP=3.9678 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,504B, BPFP=11.5977 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,512B, BPFP=3.6934 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,176B, BPFP=11.5176 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,088B, BPFP=3.4844 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,020B, BPFP=10.5029 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 184,540B, BPFP=2.2669 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 159, 4096]) +⌛️ [3/4] BACKEND: Backend time: 1.263s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 159, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000260 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000038 0.00007166 + layer.1.conv_state 0.00048847 0.40168241 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00014206 0.07085436 + layer.3.ssm_state 0.00000001 0.00000508 + layer.3.conv_state 0.00011550 0.06302404 + layer.4.ssm_state 0.00000002 0.00000613 + layer.4.conv_state 0.00021972 0.14599569 + layer.4.output 0.00000203 0.00043557 + ------------------------------------------------------------------------------------- + TOTAL 0.00002571 0.01830966 + (elements=1,470,464) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1470464 +Total Bytes 701456 +BPFP 3.8162 bits/point +EBPFP 6.6285 equivalent bits/point +MSE 0.018310 +---------------------- -------------------------------------------------------- +Time: 1.705s Load: 0.008s, Pack+Encode: 0.434s, Decode+Unpack: 1.263s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 159, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0183 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample190-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample190-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample2-layer4-item1.zst (38/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample2-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 310, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.016s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 310, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 310, 4096]) -> torch.Size([1, 1, 310, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,652B, BPFP=3.8850 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,084B, BPFP=2.9348 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,084B, BPFP=10.2744 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,048B, BPFP=3.9702 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,580B, BPFP=11.6162 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,424B, BPFP=3.6880 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,260B, BPFP=11.5381 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,528B, BPFP=3.4502 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,052B, BPFP=10.5107 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 327,752B, BPFP=2.0650 +⌛️ [2/4] FRONTEND: Frontend time: 0.580s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 310, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.943s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 310, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000256 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007352 + layer.1.conv_state 0.00048398 0.40502703 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00013032 0.07053283 + layer.3.ssm_state 0.00000001 0.00000508 + layer.3.conv_state 0.00007314 0.06456454 + layer.4.ssm_state 0.00000002 0.00000654 + layer.4.conv_state 0.00041555 0.16384108 + layer.4.output 0.00000121 0.00027057 + ------------------------------------------------------------------------------------- + TOTAL 0.00002034 0.01326888 + (elements=2,088,960) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 2088960 +Total Bytes 844088 +BPFP 3.2326 bits/point +EBPFP 5.2100 equivalent bits/point +MSE 0.013269 +---------------------- -------------------------------------------------------- +Time: 1.540s Load: 0.016s, Pack+Encode: 0.580s, Decode+Unpack: 0.943s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 310, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0133 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample2-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample2-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample20-layer4-item1.zst (39/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample20-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 194, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 194, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 194, 4096]) -> torch.Size([1, 1, 194, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,628B, BPFP=3.8835 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,324B, BPFP=2.9495 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,096B, BPFP=10.2773 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,888B, BPFP=3.9604 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,600B, BPFP=11.6211 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,804B, BPFP=3.7112 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,188B, BPFP=11.5205 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,472B, BPFP=3.4468 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,680B, BPFP=10.4199 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 222,380B, BPFP=2.2388 +⌛️ [2/4] FRONTEND: Frontend time: 0.457s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 194, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.882s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 194, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007216 + layer.1.conv_state 0.00048732 0.40460986 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00011647 0.07025528 + layer.3.ssm_state 0.00000001 0.00000507 + layer.3.conv_state 0.00011908 0.06198033 + layer.4.ssm_state 0.00000001 0.00000633 + layer.4.conv_state 0.00022797 0.15218972 + layer.4.output 0.00000187 0.00039696 + ------------------------------------------------------------------------------------- + TOTAL 0.00002322 0.01685475 + (elements=1,613,824) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1613824 +Total Bytes 738684 +BPFP 3.6618 bits/point +EBPFP 6.2212 equivalent bits/point +MSE 0.016855 +---------------------- -------------------------------------------------------- +Time: 1.348s Load: 0.009s, Pack+Encode: 0.457s, Decode+Unpack: 0.882s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 194, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0169 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample20-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample20-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample204-layer4-item1.zst (40/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample204-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 172, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 172, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 172, 4096]) -> torch.Size([1, 1, 172, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,596B, BPFP=3.8816 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,168B, BPFP=2.9399 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,064B, BPFP=10.2695 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,080B, BPFP=3.9722 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,444B, BPFP=11.5830 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,960B, BPFP=3.7207 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,340B, BPFP=11.5576 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,944B, BPFP=3.4756 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,116B, BPFP=10.5264 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 195,444B, BPFP=2.2193 +⌛️ [2/4] FRONTEND: Frontend time: 0.437s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 172, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.855s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 172, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000256 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007346 + layer.1.conv_state 0.00049938 0.40292165 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00013201 0.07039214 + layer.3.ssm_state 0.00000001 0.00000516 + layer.3.conv_state 0.00007471 0.06308520 + layer.4.ssm_state 0.00000001 0.00000642 + layer.4.conv_state 0.00021358 0.15206611 + layer.4.output 0.00000218 0.00033720 + ------------------------------------------------------------------------------------- + TOTAL 0.00002396 0.01778830 + (elements=1,523,712) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1523712 +Total Bytes 712780 +BPFP 3.7423 bits/point +EBPFP 6.4585 equivalent bits/point +MSE 0.017788 +---------------------- -------------------------------------------------------- +Time: 1.301s Load: 0.009s, Pack+Encode: 0.437s, Decode+Unpack: 0.855s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 172, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0178 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample204-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample204-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample21-layer4-item1.zst (41/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample21-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 168, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) -> torch.Size([1, 1, 168, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,520B, BPFP=3.8770 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,516B, BPFP=2.9612 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,136B, BPFP=10.2871 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,160B, BPFP=3.9771 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,652B, BPFP=11.6338 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,948B, BPFP=3.7200 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,132B, BPFP=11.5068 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,144B, BPFP=3.4268 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,836B, BPFP=10.4580 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 198,396B, BPFP=2.3065 +⌛️ [2/4] FRONTEND: Frontend time: 0.440s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.865s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000036 0.00007102 + layer.1.conv_state 0.00050661 0.40364814 + layer.2.ssm_state 0.00000001 0.00000369 + layer.2.conv_state 0.00012774 0.07086959 + layer.3.ssm_state 0.00000001 0.00000501 + layer.3.conv_state 0.00007574 0.06297226 + layer.4.ssm_state 0.00000011 0.00000726 + layer.4.conv_state 0.00082926 0.16705340 + layer.4.output 0.00000222 0.00036625 + ------------------------------------------------------------------------------------- + TOTAL 0.00003770 0.01834062 + (elements=1,507,328) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1507328 +Total Bytes 715064 +BPFP 3.7951 bits/point +EBPFP 6.5373 equivalent bits/point +MSE 0.018341 +---------------------- -------------------------------------------------------- +Time: 1.316s Load: 0.012s, Pack+Encode: 0.440s, Decode+Unpack: 0.865s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0183 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample21-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample21-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample218-layer4-item1.zst (42/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample218-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 160, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 160, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 160, 4096]) -> torch.Size([1, 1, 160, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,528B, BPFP=3.8774 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,160B, BPFP=2.9395 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,164B, BPFP=10.2939 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,928B, BPFP=3.9629 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,592B, BPFP=11.6191 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,936B, BPFP=3.7192 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,104B, BPFP=11.5000 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,508B, BPFP=3.4490 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,900B, BPFP=10.4736 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 185,344B, BPFP=2.2625 +⌛️ [2/4] FRONTEND: Frontend time: 0.443s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 160, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.860s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 160, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000030 0.00007201 + layer.1.conv_state 0.00048992 0.40384024 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00014897 0.07103047 + layer.3.ssm_state 0.00000001 0.00000506 + layer.3.conv_state 0.00007436 0.06293020 + layer.4.ssm_state 0.00000001 0.00000664 + layer.4.conv_state 0.00026192 0.15242152 + layer.4.output 0.00000216 0.00039916 + ------------------------------------------------------------------------------------- + TOTAL 0.00002591 0.01843646 + (elements=1,474,560) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1474560 +Total Bytes 701788 +BPFP 3.8074 bits/point +EBPFP 6.6093 equivalent bits/point +MSE 0.018436 +---------------------- -------------------------------------------------------- +Time: 1.312s Load: 0.009s, Pack+Encode: 0.443s, Decode+Unpack: 0.860s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 160, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0184 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample218-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample218-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample22-layer4-item1.zst (43/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample22-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 184, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 184, 4096]) -> torch.Size([1, 1, 184, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,380B, BPFP=3.8684 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,368B, BPFP=2.9521 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,260B, BPFP=10.3174 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,944B, BPFP=3.9639 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,492B, BPFP=11.5947 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,308B, BPFP=3.6809 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,124B, BPFP=11.5049 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,392B, BPFP=3.4419 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,612B, BPFP=10.4033 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 205,964B, BPFP=2.1863 +⌛️ [2/4] FRONTEND: Frontend time: 0.439s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 184, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.863s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 184, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007223 + layer.1.conv_state 0.00050597 0.40733343 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00012948 0.07103443 + layer.3.ssm_state 0.00000001 0.00000485 + layer.3.conv_state 0.00007506 0.06243151 + layer.4.ssm_state 0.00000001 0.00000632 + layer.4.conv_state 0.00035639 0.15549050 + layer.4.output 0.00000208 0.00034898 + ------------------------------------------------------------------------------------- + TOTAL 0.00002628 0.01741147 + (elements=1,572,864) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1572864 +Total Bytes 721468 +BPFP 3.6696 bits/point +EBPFP 6.2916 equivalent bits/point +MSE 0.017411 +---------------------- -------------------------------------------------------- +Time: 1.312s Load: 0.009s, Pack+Encode: 0.439s, Decode+Unpack: 0.863s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0174 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample22-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample22-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample23-layer4-item1.zst (44/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample23-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 183, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 183, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) -> torch.Size([1, 1, 183, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,384B, BPFP=3.8687 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,408B, BPFP=2.9546 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,932B, BPFP=3.9631 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,504B, BPFP=11.5977 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,272B, BPFP=3.6787 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,172B, BPFP=11.5166 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,108B, BPFP=3.4856 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,732B, BPFP=10.4326 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 206,752B, BPFP=2.2066 +⌛️ [2/4] FRONTEND: Frontend time: 0.437s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.878s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007287 + layer.1.conv_state 0.00049328 0.40598997 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00013895 0.07126924 + layer.3.ssm_state 0.00000001 0.00000475 + layer.3.conv_state 0.00011983 0.06276500 + layer.4.ssm_state 0.00000001 0.00000603 + layer.4.conv_state 0.00020516 0.14646588 + layer.4.output 0.00000202 0.00043707 + ------------------------------------------------------------------------------------- + TOTAL 0.00002402 0.01729344 + (elements=1,568,768) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1568768 +Total Bytes 723096 +BPFP 3.6875 bits/point +EBPFP 6.3206 equivalent bits/point +MSE 0.017293 +---------------------- -------------------------------------------------------- +Time: 1.327s Load: 0.012s, Pack+Encode: 0.437s, Decode+Unpack: 0.878s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 183, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0173 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample23-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample23-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample234-layer4-item1.zst (45/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample234-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 164, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 164, 4096]) -> torch.Size([1, 1, 164, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,672B, BPFP=3.8862 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,392B, BPFP=2.9536 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,148B, BPFP=10.2900 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,948B, BPFP=3.9641 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,528B, BPFP=11.6035 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,568B, BPFP=3.6968 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,268B, BPFP=11.5400 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,996B, BPFP=3.4788 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,180B, BPFP=10.5420 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 190,432B, BPFP=2.2679 +⌛️ [2/4] FRONTEND: Frontend time: 0.445s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 164, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.863s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 164, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007225 + layer.1.conv_state 0.00051090 0.40292999 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00012582 0.07062786 + layer.3.ssm_state 0.00000001 0.00000501 + layer.3.conv_state 0.00007785 0.06297633 + layer.4.ssm_state 0.00000001 0.00000657 + layer.4.conv_state 0.00021500 0.15050852 + layer.4.output 0.00000225 0.00036819 + ------------------------------------------------------------------------------------- + TOTAL 0.00002468 0.01815443 + (elements=1,490,944) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1490944 +Total Bytes 707756 +BPFP 3.7976 bits/point +EBPFP 6.5734 equivalent bits/point +MSE 0.018154 +---------------------- -------------------------------------------------------- +Time: 1.318s Load: 0.009s, Pack+Encode: 0.445s, Decode+Unpack: 0.863s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0182 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample234-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample234-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample24-layer4-item1.zst (46/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample24-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 196, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.013s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 196, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 196, 4096]) -> torch.Size([1, 1, 196, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,704B, BPFP=3.8882 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,428B, BPFP=2.9558 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,128B, BPFP=10.2852 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,976B, BPFP=3.9658 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,520B, BPFP=11.6016 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,804B, BPFP=3.7112 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,176B, BPFP=11.5176 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,592B, BPFP=3.4541 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,900B, BPFP=10.4736 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 216,108B, BPFP=2.1535 +⌛️ [2/4] FRONTEND: Frontend time: 0.453s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 196, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.904s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 196, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007271 + layer.1.conv_state 0.00048758 0.40426302 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00011566 0.06994780 + layer.3.ssm_state 0.00000001 0.00000484 + layer.3.conv_state 0.00011820 0.06235097 + layer.4.ssm_state 0.00000001 0.00000621 + layer.4.conv_state 0.00025975 0.15411876 + layer.4.output 0.00000193 0.00030719 + ------------------------------------------------------------------------------------- + TOTAL 0.00002375 0.01676045 + (elements=1,622,016) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1622016 +Total Bytes 732960 +BPFP 3.6151 bits/point +EBPFP 6.1642 equivalent bits/point +MSE 0.016760 +---------------------- -------------------------------------------------------- +Time: 1.370s Load: 0.013s, Pack+Encode: 0.453s, Decode+Unpack: 0.904s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 196, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0168 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample24-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample24-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample25-layer4-item1.zst (47/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample25-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 202, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 202, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 202, 4096]) -> torch.Size([1, 1, 202, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,580B, BPFP=3.8806 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,380B, BPFP=2.9529 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,116B, BPFP=10.2822 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,788B, BPFP=3.9543 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,472B, BPFP=11.5898 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,656B, BPFP=3.7021 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,048B, BPFP=11.4863 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,836B, BPFP=3.4690 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,732B, BPFP=10.4326 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 228,692B, BPFP=2.2112 +⌛️ [2/4] FRONTEND: Frontend time: 0.466s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 202, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.880s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 202, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007225 + layer.1.conv_state 0.00050439 0.39957076 + layer.2.ssm_state 0.00000001 0.00000358 + layer.2.conv_state 0.00014180 0.06975631 + layer.3.ssm_state 0.00000001 0.00000490 + layer.3.conv_state 0.00006834 0.06129836 + layer.4.ssm_state 0.00000001 0.00000592 + layer.4.conv_state 0.00020995 0.14460826 + layer.4.output 0.00000183 0.00035094 + ------------------------------------------------------------------------------------- + TOTAL 0.00002224 0.01622941 + (elements=1,646,592) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1646592 +Total Bytes 744924 +BPFP 3.6192 bits/point +EBPFP 6.1274 equivalent bits/point +MSE 0.016229 +---------------------- -------------------------------------------------------- +Time: 1.359s Load: 0.012s, Pack+Encode: 0.466s, Decode+Unpack: 0.880s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 202, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0162 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample25-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample25-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample26-layer4-item1.zst (48/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample26-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 196, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 196, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 196, 4096]) -> torch.Size([1, 1, 196, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,652B, BPFP=3.8850 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,540B, BPFP=2.9626 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,088B, BPFP=10.2754 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,200B, BPFP=3.9795 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,492B, BPFP=11.5947 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,612B, BPFP=3.6995 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,156B, BPFP=11.5127 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,760B, BPFP=3.4644 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,864B, BPFP=10.4648 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 219,536B, BPFP=2.1877 +⌛️ [2/4] FRONTEND: Frontend time: 0.458s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 196, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.884s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 196, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007266 + layer.1.conv_state 0.00047314 0.40524858 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00012900 0.07022273 + layer.3.ssm_state 0.00000001 0.00000498 + layer.3.conv_state 0.00011898 0.06242775 + layer.4.ssm_state 0.00000002 0.00000656 + layer.4.conv_state 0.00022845 0.14859240 + layer.4.output 0.00000185 0.00030533 + ------------------------------------------------------------------------------------- + TOTAL 0.00002308 0.01667494 + (elements=1,622,016) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1622016 +Total Bytes 736524 +BPFP 3.6326 bits/point +EBPFP 6.1825 equivalent bits/point +MSE 0.016675 +---------------------- -------------------------------------------------------- +Time: 1.351s Load: 0.009s, Pack+Encode: 0.458s, Decode+Unpack: 0.884s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 196, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0167 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample26-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample26-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample266-layer4-item1.zst (49/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample266-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 157, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 157, 4096]) -> torch.Size([1, 1, 157, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,700B, BPFP=3.8879 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,256B, BPFP=2.9453 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,096B, BPFP=10.2773 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,020B, BPFP=3.9685 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,564B, BPFP=11.6123 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,876B, BPFP=3.7156 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,376B, BPFP=11.5664 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,264B, BPFP=3.4951 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,976B, BPFP=10.4922 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 183,248B, BPFP=2.2797 +⌛️ [2/4] FRONTEND: Frontend time: 0.440s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 157, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.859s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 157, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000030 0.00007295 + layer.1.conv_state 0.00049487 0.40258425 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00012267 0.07053324 + layer.3.ssm_state 0.00000001 0.00000491 + layer.3.conv_state 0.00011738 0.06290557 + layer.4.ssm_state 0.00000002 0.00000613 + layer.4.conv_state 0.00024516 0.14461173 + layer.4.output 0.00000212 0.00039568 + ------------------------------------------------------------------------------------- + TOTAL 0.00002620 0.01837168 + (elements=1,462,272) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1462272 +Total Bytes 701000 +BPFP 3.8351 bits/point +EBPFP 6.6677 equivalent bits/point +MSE 0.018372 +---------------------- -------------------------------------------------------- +Time: 1.310s Load: 0.011s, Pack+Encode: 0.440s, Decode+Unpack: 0.859s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0184 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample266-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample266-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample27-layer4-item1.zst (50/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample27-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 195, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 195, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 195, 4096]) -> torch.Size([1, 1, 195, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,572B, BPFP=3.8801 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,192B, BPFP=2.9414 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,160B, BPFP=10.2930 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,964B, BPFP=3.9651 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,544B, BPFP=11.6074 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,536B, BPFP=3.6948 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,144B, BPFP=11.5098 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,456B, BPFP=3.4458 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,780B, BPFP=10.4443 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 221,036B, BPFP=2.2139 +⌛️ [2/4] FRONTEND: Frontend time: 0.454s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 195, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.887s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 195, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000009 0.00007271 + layer.1.conv_state 0.00049297 0.40426612 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013859 0.07075693 + layer.3.ssm_state 0.00000001 0.00000500 + layer.3.conv_state 0.00007636 0.06272336 + layer.4.ssm_state 0.00000001 0.00000633 + layer.4.conv_state 0.00030466 0.15460028 + layer.4.output 0.00000195 0.00040894 + ------------------------------------------------------------------------------------- + TOTAL 0.00002444 0.01688610 + (elements=1,617,920) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1617920 +Total Bytes 737008 +BPFP 3.6442 bits/point +EBPFP 6.1955 equivalent bits/point +MSE 0.016886 +---------------------- -------------------------------------------------------- +Time: 1.353s Load: 0.011s, Pack+Encode: 0.454s, Decode+Unpack: 0.887s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 195, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0169 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample27-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample27-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample283-layer4-item1.zst (51/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample283-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 154, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 154, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 154, 4096]) -> torch.Size([1, 1, 154, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,540B, BPFP=3.8782 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,556B, BPFP=2.9636 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,112B, BPFP=10.2812 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,976B, BPFP=3.9658 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,548B, BPFP=11.6084 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,968B, BPFP=3.7212 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,220B, BPFP=11.5283 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,940B, BPFP=3.4753 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,904B, BPFP=10.4746 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 175,284B, BPFP=2.2231 +⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 154, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.852s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 154, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000257 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000039 0.00007162 + layer.1.conv_state 0.00049616 0.40321556 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00014030 0.07094324 + layer.3.ssm_state 0.00000001 0.00000515 + layer.3.conv_state 0.00011494 0.06298051 + layer.4.ssm_state 0.00000003 0.00000610 + layer.4.conv_state 0.00023159 0.14400519 + layer.4.output 0.00000204 0.00042205 + ------------------------------------------------------------------------------------- + TOTAL 0.00002644 0.01854693 + (elements=1,449,984) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1449984 +Total Bytes 692672 +BPFP 3.8217 bits/point +EBPFP 6.6763 equivalent bits/point +MSE 0.018547 +---------------------- -------------------------------------------------------- +Time: 1.295s Load: 0.008s, Pack+Encode: 0.435s, Decode+Unpack: 0.852s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 154, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0185 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample283-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample283-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample29-layer4-item1.zst (52/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample29-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 186, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 186, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 186, 4096]) -> torch.Size([1, 1, 186, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,588B, BPFP=3.8811 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,408B, BPFP=2.9546 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,184B, BPFP=10.2988 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,136B, BPFP=3.9756 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,588B, BPFP=11.6182 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,172B, BPFP=3.7336 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,192B, BPFP=11.5215 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,032B, BPFP=3.4810 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 205,192B, BPFP=2.1547 +⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 186, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.861s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 186, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007199 + layer.1.conv_state 0.00049578 0.40586033 + layer.2.ssm_state 0.00000001 0.00000370 + layer.2.conv_state 0.00013264 0.07131118 + layer.3.ssm_state 0.00000001 0.00000503 + layer.3.conv_state 0.00012070 0.06329720 + layer.4.ssm_state 0.00000002 0.00000601 + layer.4.conv_state 0.00022222 0.14257394 + layer.4.output 0.00000189 0.00037982 + ------------------------------------------------------------------------------------- + TOTAL 0.00002409 0.01706335 + (elements=1,581,056) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1581056 +Total Bytes 722740 +BPFP 3.6570 bits/point +EBPFP 6.2757 equivalent bits/point +MSE 0.017063 +---------------------- -------------------------------------------------------- +Time: 1.308s Load: 0.012s, Pack+Encode: 0.435s, Decode+Unpack: 0.861s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 186, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0171 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample29-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample29-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample3-layer4-item1.zst (53/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample3-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 276, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 276, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 276, 4096]) -> torch.Size([1, 1, 276, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,648B, BPFP=3.8848 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,364B, BPFP=2.9519 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,104B, BPFP=10.2793 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,168B, BPFP=3.9775 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,356B, BPFP=11.5615 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,832B, BPFP=3.7129 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,132B, BPFP=11.5068 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,100B, BPFP=3.4241 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,836B, BPFP=10.4580 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 293,756B, BPFP=2.0788 +⌛️ [2/4] FRONTEND: Frontend time: 0.462s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 276, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.910s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 276, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007206 + layer.1.conv_state 0.00048777 0.40325505 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00011906 0.06971700 + layer.3.ssm_state 0.00000001 0.00000494 + layer.3.conv_state 0.00011276 0.06230633 + layer.4.ssm_state 0.00000006 0.00000673 + layer.4.conv_state 0.00059394 0.16348106 + layer.4.output 0.00000125 0.00023975 + ------------------------------------------------------------------------------------- + TOTAL 0.00002528 0.01409187 + (elements=1,949,696) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1949696 +Total Bytes 809920 +BPFP 3.3233 bits/point +EBPFP 5.4412 equivalent bits/point +MSE 0.014092 +---------------------- -------------------------------------------------------- +Time: 1.385s Load: 0.012s, Pack+Encode: 0.462s, Decode+Unpack: 0.910s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 276, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0141 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample3-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample3-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample30-layer4-item1.zst (54/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample30-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 198, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 198, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 198, 4096]) -> torch.Size([1, 1, 198, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,592B, BPFP=3.8813 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,340B, BPFP=2.9504 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,124B, BPFP=10.2842 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,808B, BPFP=3.9556 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,472B, BPFP=11.5898 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,048B, BPFP=3.7261 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,252B, BPFP=11.5361 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,760B, BPFP=3.5254 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,988B, BPFP=10.4951 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 239,532B, BPFP=2.3628 +⌛️ [2/4] FRONTEND: Frontend time: 0.452s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 198, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.888s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 198, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007142 + layer.1.conv_state 0.00048081 0.40451014 + layer.2.ssm_state 0.00000001 0.00000368 + layer.2.conv_state 0.00011915 0.07040183 + layer.3.ssm_state 0.00000001 0.00000509 + layer.3.conv_state 0.00011359 0.06293563 + layer.4.ssm_state 0.00000002 0.00000556 + layer.4.conv_state 0.00023406 0.13465719 + layer.4.output 0.00000200 0.00042771 + ------------------------------------------------------------------------------------- + TOTAL 0.00002301 0.01637226 + (elements=1,630,208) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1630208 +Total Bytes 757540 +BPFP 3.7175 bits/point +EBPFP 6.2596 equivalent bits/point +MSE 0.016372 +---------------------- -------------------------------------------------------- +Time: 1.350s Load: 0.010s, Pack+Encode: 0.452s, Decode+Unpack: 0.888s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 198, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0164 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample30-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample30-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample31-layer4-item1.zst (55/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample31-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 175, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 175, 4096]) -> torch.Size([1, 1, 175, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,652B, BPFP=3.8850 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,332B, BPFP=2.9500 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,152B, BPFP=10.2910 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,864B, BPFP=3.9590 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,580B, BPFP=11.6162 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,584B, BPFP=3.6978 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,208B, BPFP=11.5254 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,408B, BPFP=3.4429 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,788B, BPFP=10.4463 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 204,600B, BPFP=2.2835 +⌛️ [2/4] FRONTEND: Frontend time: 0.437s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 175, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.872s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 175, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007382 + layer.1.conv_state 0.00049445 0.40665838 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00015634 0.07106288 + layer.3.ssm_state 0.00000001 0.00000485 + layer.3.conv_state 0.00006751 0.06309474 + layer.4.ssm_state 0.00000001 0.00000604 + layer.4.conv_state 0.00021972 0.15023774 + layer.4.output 0.00000206 0.00042313 + ------------------------------------------------------------------------------------- + TOTAL 0.00002411 0.01774398 + (elements=1,536,000) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1536000 +Total Bytes 720792 +BPFP 3.7541 bits/point +EBPFP 6.4426 equivalent bits/point +MSE 0.017744 +---------------------- -------------------------------------------------------- +Time: 1.318s Load: 0.009s, Pack+Encode: 0.437s, Decode+Unpack: 0.872s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 175, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0177 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample31-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample31-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample32-layer4-item1.zst (56/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample32-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 225, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.013s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 225, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 225, 4096]) -> torch.Size([1, 1, 225, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,684B, BPFP=3.8870 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,508B, BPFP=2.9607 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,116B, BPFP=10.2822 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,052B, BPFP=3.9705 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,524B, BPFP=11.6025 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,000B, BPFP=3.7231 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,256B, BPFP=11.5371 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,908B, BPFP=3.4734 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,880B, BPFP=10.4688 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 256,624B, BPFP=2.2276 +⌛️ [2/4] FRONTEND: Frontend time: 0.455s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 225, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.883s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 225, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007209 + layer.1.conv_state 0.00049817 0.40178490 + layer.2.ssm_state 0.00000001 0.00000370 + layer.2.conv_state 0.00012605 0.07037306 + layer.3.ssm_state 0.00000001 0.00000516 + layer.3.conv_state 0.00007245 0.06315585 + layer.4.ssm_state 0.00000001 0.00000633 + layer.4.conv_state 0.00021821 0.14824407 + layer.4.output 0.00000156 0.00032190 + ------------------------------------------------------------------------------------- + TOTAL 0.00002082 0.01551146 + (elements=1,740,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1740800 +Total Bytes 774176 +BPFP 3.5578 bits/point +EBPFP 5.9363 equivalent bits/point +MSE 0.015511 +---------------------- -------------------------------------------------------- +Time: 1.351s Load: 0.013s, Pack+Encode: 0.455s, Decode+Unpack: 0.883s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 225, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0155 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample32-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample32-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample323-layer4-item1.zst (57/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample323-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 152, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 152, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 152, 4096]) -> torch.Size([1, 1, 152, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,564B, BPFP=3.8796 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,280B, BPFP=2.9468 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,024B, BPFP=10.2598 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,100B, BPFP=3.9734 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,440B, BPFP=11.5820 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,992B, BPFP=3.7227 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,136B, BPFP=11.5078 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,172B, BPFP=3.4895 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,900B, BPFP=10.4736 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 179,736B, BPFP=2.3095 +⌛️ [2/4] FRONTEND: Frontend time: 0.438s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 152, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.855s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 152, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007285 + layer.1.conv_state 0.00048212 0.40256870 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00013258 0.07057328 + layer.3.ssm_state 0.00000001 0.00000497 + layer.3.conv_state 0.00007465 0.06293070 + layer.4.ssm_state 0.00000001 0.00000618 + layer.4.conv_state 0.00020440 0.14720553 + layer.4.output 0.00000217 0.00038716 + ------------------------------------------------------------------------------------- + TOTAL 0.00002460 0.01868343 + (elements=1,441,792) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1441792 +Total Bytes 696968 +BPFP 3.8672 bits/point +EBPFP 6.7372 equivalent bits/point +MSE 0.018683 +---------------------- -------------------------------------------------------- +Time: 1.302s Load: 0.009s, Pack+Encode: 0.438s, Decode+Unpack: 0.855s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 152, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0187 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample323-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample323-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample33-layer4-item1.zst (58/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample33-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 205, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 205, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 205, 4096]) -> torch.Size([1, 1, 205, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,588B, BPFP=3.8811 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,396B, BPFP=2.9539 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,168B, BPFP=10.2949 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,132B, BPFP=3.9753 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,388B, BPFP=11.5693 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,356B, BPFP=3.6838 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,056B, BPFP=11.4883 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,184B, BPFP=3.4902 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,716B, BPFP=10.4287 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 225,684B, BPFP=2.1502 +⌛️ [2/4] FRONTEND: Frontend time: 0.456s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 205, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.886s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 205, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000024 0.00007139 + layer.1.conv_state 0.00049079 0.39976415 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00014720 0.06932274 + layer.3.ssm_state 0.00000001 0.00000503 + layer.3.conv_state 0.00011227 0.06187543 + layer.4.ssm_state 0.00000001 0.00000616 + layer.4.conv_state 0.00021844 0.14942203 + layer.4.output 0.00000170 0.00029112 + ------------------------------------------------------------------------------------- + TOTAL 0.00002290 0.01618321 + (elements=1,658,880) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1658880 +Total Bytes 742292 +BPFP 3.5797 bits/point +EBPFP 6.0711 equivalent bits/point +MSE 0.016183 +---------------------- -------------------------------------------------------- +Time: 1.354s Load: 0.012s, Pack+Encode: 0.456s, Decode+Unpack: 0.886s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 205, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0162 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample33-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample33-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample34-layer4-item1.zst (59/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample34-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 223, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 223, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 223, 4096]) -> torch.Size([1, 1, 223, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,580B, BPFP=3.8806 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,028B, BPFP=2.9314 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,156B, BPFP=10.2920 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,044B, BPFP=3.9700 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,484B, BPFP=11.5928 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,696B, BPFP=3.7046 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,164B, BPFP=11.5146 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,472B, BPFP=3.4468 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,116B, BPFP=10.5264 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 245,292B, BPFP=2.1484 +⌛️ [2/4] FRONTEND: Frontend time: 0.447s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 223, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.911s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 223, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007226 + layer.1.conv_state 0.00048351 0.40235260 + layer.2.ssm_state 0.00000001 0.00000358 + layer.2.conv_state 0.00011966 0.06991211 + layer.3.ssm_state 0.00000001 0.00000507 + layer.3.conv_state 0.00006977 0.06301930 + layer.4.ssm_state 0.00000002 0.00000688 + layer.4.conv_state 0.00033313 0.16363548 + layer.4.output 0.00000142 0.00032292 + ------------------------------------------------------------------------------------- + TOTAL 0.00002256 0.01587439 + (elements=1,732,608) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1732608 +Total Bytes 761656 +BPFP 3.5168 bits/point +EBPFP 5.9010 equivalent bits/point +MSE 0.015874 +---------------------- -------------------------------------------------------- +Time: 1.369s Load: 0.010s, Pack+Encode: 0.447s, Decode+Unpack: 0.911s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 223, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0159 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample34-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample34-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample35-layer4-item1.zst (60/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample35-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 183, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 183, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) -> torch.Size([1, 1, 183, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,660B, BPFP=3.8855 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,384B, BPFP=2.9531 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,080B, BPFP=10.2734 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,284B, BPFP=3.9846 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,572B, BPFP=11.6143 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,740B, BPFP=3.7073 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,196B, BPFP=11.5225 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,940B, BPFP=3.4753 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,940B, BPFP=10.4834 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 200,772B, BPFP=2.1428 +⌛️ [2/4] FRONTEND: Frontend time: 0.441s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.848s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000021 0.00007095 + layer.1.conv_state 0.00049749 0.40290466 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00012539 0.07035200 + layer.3.ssm_state 0.00000001 0.00000497 + layer.3.conv_state 0.00007456 0.06263165 + layer.4.ssm_state 0.00000001 0.00000654 + layer.4.conv_state 0.00032004 0.15521936 + layer.4.output 0.00000203 0.00040406 + ------------------------------------------------------------------------------------- + TOTAL 0.00002530 0.01737401 + (elements=1,568,768) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1568768 +Total Bytes 718192 +BPFP 3.6625 bits/point +EBPFP 6.3011 equivalent bits/point +MSE 0.017374 +---------------------- -------------------------------------------------------- +Time: 1.299s Load: 0.010s, Pack+Encode: 0.441s, Decode+Unpack: 0.848s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 183, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0174 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample35-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample35-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample36-layer4-item1.zst (61/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample36-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 181, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 181, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 181, 4096]) -> torch.Size([1, 1, 181, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,620B, BPFP=3.8831 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,360B, BPFP=2.9517 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,164B, BPFP=10.2939 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,980B, BPFP=3.9661 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,612B, BPFP=11.6240 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,756B, BPFP=3.7083 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,300B, BPFP=11.5479 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,652B, BPFP=3.4578 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,108B, BPFP=10.5244 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 208,012B, BPFP=2.2446 +⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 181, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.851s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 181, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007230 + layer.1.conv_state 0.00048788 0.40508354 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00012276 0.07033268 + layer.3.ssm_state 0.00000001 0.00000516 + layer.3.conv_state 0.00007521 0.06390835 + layer.4.ssm_state 0.00000001 0.00000643 + layer.4.conv_state 0.00025615 0.15649608 + layer.4.output 0.00000200 0.00034955 + ------------------------------------------------------------------------------------- + TOTAL 0.00002381 0.01753628 + (elements=1,560,576) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1560576 +Total Bytes 725188 +BPFP 3.7175 bits/point +EBPFP 6.3687 equivalent bits/point +MSE 0.017536 +---------------------- -------------------------------------------------------- +Time: 1.298s Load: 0.012s, Pack+Encode: 0.435s, Decode+Unpack: 0.851s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 181, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0175 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample36-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample36-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample37-layer4-item1.zst (62/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample37-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 204, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 204, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 204, 4096]) -> torch.Size([1, 1, 204, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,692B, BPFP=3.8875 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,408B, BPFP=2.9546 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,112B, BPFP=10.2812 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,144B, BPFP=3.9761 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,504B, BPFP=11.5977 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,080B, BPFP=3.7280 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,264B, BPFP=11.5391 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,188B, BPFP=3.4294 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,088B, BPFP=10.5195 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 234,704B, BPFP=2.2471 +⌛️ [2/4] FRONTEND: Frontend time: 0.444s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 204, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.877s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 204, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007249 + layer.1.conv_state 0.00048573 0.40615347 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00014000 0.07056384 + layer.3.ssm_state 0.00000001 0.00000512 + layer.3.conv_state 0.00007101 0.06322234 + layer.4.ssm_state 0.00000002 0.00000670 + layer.4.conv_state 0.00054851 0.16280091 + layer.4.output 0.00000173 0.00033265 + ------------------------------------------------------------------------------------- + TOTAL 0.00002845 0.01668636 + (elements=1,654,784) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1654784 +Total Bytes 751808 +BPFP 3.6346 bits/point +EBPFP 6.1345 equivalent bits/point +MSE 0.016686 +---------------------- -------------------------------------------------------- +Time: 1.331s Load: 0.010s, Pack+Encode: 0.444s, Decode+Unpack: 0.877s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 204, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0167 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample37-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample37-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample38-layer4-item1.zst (63/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample38-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 189, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 189, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 189, 4096]) -> torch.Size([1, 1, 189, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,644B, BPFP=3.8845 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,468B, BPFP=2.9583 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,148B, BPFP=10.2900 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,968B, BPFP=3.9653 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,524B, BPFP=11.6025 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,688B, BPFP=3.7041 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,228B, BPFP=11.5303 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,904B, BPFP=3.4731 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,000B, BPFP=10.4980 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 202,972B, BPFP=2.0975 +⌛️ [2/4] FRONTEND: Frontend time: 0.441s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 189, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.865s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 189, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000256 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007263 + layer.1.conv_state 0.00048943 0.40273392 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00011993 0.07049067 + layer.3.ssm_state 0.00000001 0.00000501 + layer.3.conv_state 0.00011634 0.06245555 + layer.4.ssm_state 0.00000001 0.00000624 + layer.4.conv_state 0.00025516 0.15170226 + layer.4.output 0.00000196 0.00031197 + ------------------------------------------------------------------------------------- + TOTAL 0.00002414 0.01699103 + (elements=1,593,344) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1593344 +Total Bytes 720168 +BPFP 3.6159 bits/point +EBPFP 6.2127 equivalent bits/point +MSE 0.016991 +---------------------- -------------------------------------------------------- +Time: 1.316s Load: 0.010s, Pack+Encode: 0.441s, Decode+Unpack: 0.865s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 189, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0170 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample38-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample38-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample39-layer4-item1.zst (64/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample39-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 188, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 188, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 188, 4096]) -> torch.Size([1, 1, 188, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,512B, BPFP=3.8765 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,404B, BPFP=2.9543 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,112B, BPFP=10.2812 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,940B, BPFP=3.9636 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,504B, BPFP=11.5977 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,060B, BPFP=3.7268 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,156B, BPFP=11.5127 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,528B, BPFP=3.5112 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,980B, BPFP=10.4932 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 206,756B, BPFP=2.1480 +⌛️ [2/4] FRONTEND: Frontend time: 0.442s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 188, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.861s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 188, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000255 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007250 + layer.1.conv_state 0.00048897 0.40450150 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00011975 0.07093962 + layer.3.ssm_state 0.00000001 0.00000493 + layer.3.conv_state 0.00006970 0.06293868 + layer.4.ssm_state 0.00000001 0.00000575 + layer.4.conv_state 0.00027957 0.13964418 + layer.4.output 0.00000196 0.00032109 + ------------------------------------------------------------------------------------- + TOTAL 0.00002374 0.01684542 + (elements=1,589,248) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1589248 +Total Bytes 724576 +BPFP 3.6474 bits/point +EBPFP 6.2540 equivalent bits/point +MSE 0.016845 +---------------------- -------------------------------------------------------- +Time: 1.313s Load: 0.010s, Pack+Encode: 0.442s, Decode+Unpack: 0.861s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 188, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0168 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample39-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample39-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample4-layer4-item1.zst (65/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample4-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 271, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.015s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 271, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 271, 4096]) -> torch.Size([1, 1, 271, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,632B, BPFP=3.8838 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,212B, BPFP=2.9426 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,148B, BPFP=10.2900 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,060B, BPFP=3.9709 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,536B, BPFP=11.6055 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,980B, BPFP=3.7219 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 46,956B, BPFP=11.4639 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,948B, BPFP=3.4758 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,844B, BPFP=10.4600 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 302,700B, BPFP=2.1816 +⌛️ [2/4] FRONTEND: Frontend time: 0.469s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 271, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.920s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 271, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007261 + layer.1.conv_state 0.00049489 0.40615779 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013525 0.07006691 + layer.3.ssm_state 0.00000001 0.00000505 + layer.3.conv_state 0.00011705 0.06223079 + layer.4.ssm_state 0.00000001 0.00000626 + layer.4.conv_state 0.00022610 0.14847855 + layer.4.output 0.00000138 0.00029359 + ------------------------------------------------------------------------------------- + TOTAL 0.00001983 0.01406905 + (elements=1,929,216) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1929216 +Total Bytes 819640 +BPFP 3.3989 bits/point +EBPFP 5.5425 equivalent bits/point +MSE 0.014069 +---------------------- -------------------------------------------------------- +Time: 1.403s Load: 0.015s, Pack+Encode: 0.469s, Decode+Unpack: 0.920s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 271, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0141 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample4-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample4-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample40-layer4-item1.zst (66/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample40-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 168, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) -> torch.Size([1, 1, 168, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,580B, BPFP=3.8806 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,448B, BPFP=2.9570 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,180B, BPFP=10.2979 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,180B, BPFP=3.9783 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,496B, BPFP=11.5957 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,736B, BPFP=3.7070 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,212B, BPFP=11.5264 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,188B, BPFP=3.4905 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,864B, BPFP=10.4648 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 200,512B, BPFP=2.3311 +⌛️ [2/4] FRONTEND: Frontend time: 0.441s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.855s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 168, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000047 0.00007418 + layer.1.conv_state 0.00050445 0.40032348 + layer.2.ssm_state 0.00000001 0.00000368 + layer.2.conv_state 0.00012625 0.07075401 + layer.3.ssm_state 0.00000001 0.00000496 + layer.3.conv_state 0.00011525 0.06274790 + layer.4.ssm_state 0.00000001 0.00000601 + layer.4.conv_state 0.00022432 0.14360315 + layer.4.output 0.00000216 0.00038198 + ------------------------------------------------------------------------------------- + TOTAL 0.00002530 0.01775851 + (elements=1,507,328) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1507328 +Total Bytes 718020 +BPFP 3.8108 bits/point +EBPFP 6.5574 equivalent bits/point +MSE 0.017759 +---------------------- -------------------------------------------------------- +Time: 1.305s Load: 0.009s, Pack+Encode: 0.441s, Decode+Unpack: 0.855s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 168, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0178 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample40-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample40-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample403-layer4-item1.zst (67/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample403-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 173, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 173, 4096]) -> torch.Size([1, 1, 173, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,564B, BPFP=3.8796 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,560B, BPFP=2.9639 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,116B, BPFP=10.2822 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,960B, BPFP=3.9648 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,600B, BPFP=11.6211 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,856B, BPFP=3.7144 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,172B, BPFP=11.5166 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,172B, BPFP=3.4285 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,908B, BPFP=10.4756 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 198,796B, BPFP=2.2444 +⌛️ [2/4] FRONTEND: Frontend time: 0.432s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 173, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.851s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 173, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007342 + layer.1.conv_state 0.00048463 0.40343171 + layer.2.ssm_state 0.00000001 0.00000368 + layer.2.conv_state 0.00012632 0.07046933 + layer.3.ssm_state 0.00000001 0.00000518 + layer.3.conv_state 0.00007282 0.06262308 + layer.4.ssm_state 0.00000003 0.00000674 + layer.4.conv_state 0.00058008 0.16601269 + layer.4.output 0.00000205 0.00034224 + ------------------------------------------------------------------------------------- + TOTAL 0.00003121 0.01804568 + (elements=1,527,808) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1527808 +Total Bytes 715328 +BPFP 3.7456 bits/point +EBPFP 6.4503 equivalent bits/point +MSE 0.018046 +---------------------- -------------------------------------------------------- +Time: 1.295s Load: 0.012s, Pack+Encode: 0.432s, Decode+Unpack: 0.851s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0180 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample403-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample403-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample41-layer4-item1.zst (68/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample41-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 197, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.013s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 197, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 197, 4096]) -> torch.Size([1, 1, 197, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,512B, BPFP=3.8765 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,428B, BPFP=2.9558 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,108B, BPFP=10.2803 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,004B, BPFP=3.9675 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,460B, BPFP=11.5869 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,620B, BPFP=3.7000 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,284B, BPFP=11.5439 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,720B, BPFP=3.4619 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,068B, BPFP=10.5146 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 225,040B, BPFP=2.2311 +⌛️ [2/4] FRONTEND: Frontend time: 0.451s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 197, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.905s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 197, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000255 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007366 + layer.1.conv_state 0.00050234 0.40298811 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00013726 0.07044253 + layer.3.ssm_state 0.00000001 0.00000500 + layer.3.conv_state 0.00007731 0.06334955 + layer.4.ssm_state 0.00000001 0.00000659 + layer.4.conv_state 0.00032633 0.15585165 + layer.4.output 0.00000190 0.00031467 + ------------------------------------------------------------------------------------- + TOTAL 0.00002493 0.01676216 + (elements=1,626,112) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1626112 +Total Bytes 741868 +BPFP 3.6498 bits/point +EBPFP 6.1924 equivalent bits/point +MSE 0.016762 +---------------------- -------------------------------------------------------- +Time: 1.370s Load: 0.013s, Pack+Encode: 0.451s, Decode+Unpack: 0.905s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 197, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0168 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample41-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample41-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample42-layer4-item1.zst (69/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample42-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 185, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 185, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 185, 4096]) -> torch.Size([1, 1, 185, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,472B, BPFP=3.8740 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,240B, BPFP=2.9443 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,072B, BPFP=10.2715 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,080B, BPFP=3.9722 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,516B, BPFP=11.6006 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,528B, BPFP=3.6943 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,148B, BPFP=11.5107 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,756B, BPFP=3.4641 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,732B, BPFP=10.4326 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 205,752B, BPFP=2.1722 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 185, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.853s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 185, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000255 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007243 + layer.1.conv_state 0.00049834 0.40453827 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00012928 0.07054266 + layer.3.ssm_state 0.00000001 0.00000494 + layer.3.conv_state 0.00007819 0.06288388 + layer.4.ssm_state 0.00000001 0.00000661 + layer.4.conv_state 0.00030927 0.15517242 + layer.4.output 0.00000194 0.00032226 + ------------------------------------------------------------------------------------- + TOTAL 0.00002507 0.01728886 + (elements=1,576,960) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1576960 +Total Bytes 721920 +BPFP 3.6623 bits/point +EBPFP 6.2809 equivalent bits/point +MSE 0.017289 +---------------------- -------------------------------------------------------- +Time: 1.299s Load: 0.012s, Pack+Encode: 0.434s, Decode+Unpack: 0.853s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 185, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0173 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample42-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample42-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample44-layer4-item1.zst (70/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample44-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 210, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.013s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 210, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 210, 4096]) -> torch.Size([1, 1, 210, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,732B, BPFP=3.8899 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,372B, BPFP=2.9524 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,084B, BPFP=10.2744 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,992B, BPFP=3.9668 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,504B, BPFP=11.5977 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,580B, BPFP=3.6975 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,340B, BPFP=11.5576 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,224B, BPFP=3.4927 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,032B, BPFP=10.5059 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 238,632B, BPFP=2.2194 +⌛️ [2/4] FRONTEND: Frontend time: 0.450s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 210, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.882s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 210, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007333 + layer.1.conv_state 0.00049741 0.40497714 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00015034 0.07065255 + layer.3.ssm_state 0.00000001 0.00000495 + layer.3.conv_state 0.00007531 0.06335539 + layer.4.ssm_state 0.00000004 0.00000625 + layer.4.conv_state 0.00020482 0.15031154 + layer.4.output 0.00000169 0.00036060 + ------------------------------------------------------------------------------------- + TOTAL 0.00002184 0.01619904 + (elements=1,679,360) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1679360 +Total Bytes 756116 +BPFP 3.6019 bits/point +EBPFP 6.0671 equivalent bits/point +MSE 0.016199 +---------------------- -------------------------------------------------------- +Time: 1.345s Load: 0.013s, Pack+Encode: 0.450s, Decode+Unpack: 0.882s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 210, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0162 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample44-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample44-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample45-layer4-item1.zst (71/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample45-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 184, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 184, 4096]) -> torch.Size([1, 1, 184, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,540B, BPFP=3.8782 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,300B, BPFP=2.9480 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,132B, BPFP=10.2861 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,832B, BPFP=3.9570 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,580B, BPFP=11.6162 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,000B, BPFP=3.7231 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,164B, BPFP=11.5146 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,832B, BPFP=3.4688 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,940B, BPFP=10.4834 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 206,128B, BPFP=2.1880 +⌛️ [2/4] FRONTEND: Frontend time: 0.433s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 184, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.851s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 184, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007201 + layer.1.conv_state 0.00051907 0.40297601 + layer.2.ssm_state 0.00000001 0.00000368 + layer.2.conv_state 0.00017046 0.07146997 + layer.3.ssm_state 0.00000001 0.00000517 + layer.3.conv_state 0.00006998 0.06325621 + layer.4.ssm_state 0.00000001 0.00000635 + layer.4.conv_state 0.00024001 0.15029754 + layer.4.output 0.00000196 0.00034417 + ------------------------------------------------------------------------------------- + TOTAL 0.00002482 0.01723647 + (elements=1,572,864) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1572864 +Total Bytes 723072 +BPFP 3.6777 bits/point +EBPFP 6.3070 equivalent bits/point +MSE 0.017236 +---------------------- -------------------------------------------------------- +Time: 1.293s Load: 0.009s, Pack+Encode: 0.433s, Decode+Unpack: 0.851s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0172 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample45-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample45-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample46-layer4-item1.zst (72/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample46-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 196, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 196, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 196, 4096]) -> torch.Size([1, 1, 196, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,624B, BPFP=3.8833 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,304B, BPFP=2.9482 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,148B, BPFP=10.2900 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,976B, BPFP=3.9658 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,540B, BPFP=11.6064 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,492B, BPFP=3.6921 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,076B, BPFP=11.4932 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,636B, BPFP=3.4568 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,732B, BPFP=10.4326 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 222,696B, BPFP=2.2191 +⌛️ [2/4] FRONTEND: Frontend time: 0.451s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 196, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.879s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 196, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007202 + layer.1.conv_state 0.00047987 0.40505731 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00014269 0.07026906 + layer.3.ssm_state 0.00000001 0.00000487 + layer.3.conv_state 0.00007044 0.06305342 + layer.4.ssm_state 0.00000001 0.00000611 + layer.4.conv_state 0.00031684 0.15634343 + layer.4.output 0.00000179 0.00031778 + ------------------------------------------------------------------------------------- + TOTAL 0.00002426 0.01684730 + (elements=1,622,016) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1622016 +Total Bytes 738848 +BPFP 3.6441 bits/point +EBPFP 6.1898 equivalent bits/point +MSE 0.016847 +---------------------- -------------------------------------------------------- +Time: 1.339s Load: 0.010s, Pack+Encode: 0.451s, Decode+Unpack: 0.879s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 196, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0168 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample46-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample46-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample47-layer4-item1.zst (73/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample47-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 203, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.013s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 203, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 203, 4096]) -> torch.Size([1, 1, 203, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,492B, BPFP=3.8752 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,360B, BPFP=2.9517 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,120B, BPFP=10.2832 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,904B, BPFP=3.9614 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,536B, BPFP=11.6055 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,844B, BPFP=3.7136 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,296B, BPFP=11.5469 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,328B, BPFP=3.4990 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,868B, BPFP=10.4658 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 233,248B, BPFP=2.2442 +⌛️ [2/4] FRONTEND: Frontend time: 0.447s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 203, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.888s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 203, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000043 0.00007193 + layer.1.conv_state 0.00048147 0.40180668 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00012906 0.07106038 + layer.3.ssm_state 0.00000001 0.00000506 + layer.3.conv_state 0.00011463 0.06351107 + layer.4.ssm_state 0.00000002 0.00000603 + layer.4.conv_state 0.00022714 0.14536370 + layer.4.output 0.00000188 0.00037364 + ------------------------------------------------------------------------------------- + TOTAL 0.00002279 0.01633063 + (elements=1,650,688) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1650688 +Total Bytes 750620 +BPFP 3.6379 bits/point +EBPFP 6.1453 equivalent bits/point +MSE 0.016331 +---------------------- -------------------------------------------------------- +Time: 1.348s Load: 0.013s, Pack+Encode: 0.447s, Decode+Unpack: 0.888s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 203, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0163 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample47-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample47-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample48-layer4-item1.zst (74/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample48-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 184, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 184, 4096]) -> torch.Size([1, 1, 184, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,540B, BPFP=3.8782 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,344B, BPFP=2.9507 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,108B, BPFP=10.2803 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,888B, BPFP=3.9604 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,536B, BPFP=11.6055 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,648B, BPFP=3.7017 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,204B, BPFP=11.5244 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,472B, BPFP=3.4468 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,828B, BPFP=10.4561 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 204,060B, BPFP=2.1661 +⌛️ [2/4] FRONTEND: Frontend time: 0.433s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 184, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.854s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 184, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007229 + layer.1.conv_state 0.00050109 0.40284425 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00013610 0.07097637 + layer.3.ssm_state 0.00000001 0.00000500 + layer.3.conv_state 0.00007718 0.06324465 + layer.4.ssm_state 0.00000001 0.00000624 + layer.4.conv_state 0.00025147 0.15122327 + layer.4.output 0.00000198 0.00033620 + ------------------------------------------------------------------------------------- + TOTAL 0.00002414 0.01723866 + (elements=1,572,864) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1572864 +Total Bytes 720252 +BPFP 3.6634 bits/point +EBPFP 6.2889 equivalent bits/point +MSE 0.017239 +---------------------- -------------------------------------------------------- +Time: 1.296s Load: 0.009s, Pack+Encode: 0.433s, Decode+Unpack: 0.854s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 184, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0172 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample48-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample48-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample51-layer4-item1.zst (75/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample51-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 192, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 192, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 192, 4096]) -> torch.Size([1, 1, 192, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,696B, BPFP=3.8877 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,272B, BPFP=2.9463 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,164B, BPFP=10.2939 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,944B, BPFP=3.9639 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,464B, BPFP=11.5879 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,940B, BPFP=3.7195 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,088B, BPFP=11.4961 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,300B, BPFP=3.4973 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,804B, BPFP=10.4502 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 202,744B, BPFP=2.0624 +⌛️ [2/4] FRONTEND: Frontend time: 0.478s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 192, 4096]) +⌛️ [3/4] BACKEND: Backend time: 1.004s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 192, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000024 0.00007403 + layer.1.conv_state 0.00048279 0.40422958 + layer.2.ssm_state 0.00000001 0.00000368 + layer.2.conv_state 0.00015517 0.07025753 + layer.3.ssm_state 0.00000001 0.00000493 + layer.3.conv_state 0.00011498 0.06161559 + layer.4.ssm_state 0.00000001 0.00000560 + layer.4.conv_state 0.00025157 0.13458523 + layer.4.output 0.00000201 0.00029954 + ------------------------------------------------------------------------------------- + TOTAL 0.00002449 0.01651665 + (elements=1,605,632) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1605632 +Total Bytes 720040 +BPFP 3.5876 bits/point +EBPFP 6.1650 equivalent bits/point +MSE 0.016517 +---------------------- -------------------------------------------------------- +Time: 1.492s Load: 0.010s, Pack+Encode: 0.478s, Decode+Unpack: 1.004s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 192, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0165 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample51-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample51-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample52-layer4-item1.zst (76/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample52-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 170, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 170, 4096]) -> torch.Size([1, 1, 170, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,596B, BPFP=3.8816 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,432B, BPFP=2.9561 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,116B, BPFP=10.2822 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,980B, BPFP=3.9661 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,476B, BPFP=11.5908 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,704B, BPFP=3.7051 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,084B, BPFP=11.4951 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,908B, BPFP=3.4734 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,944B, BPFP=10.4844 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 201,288B, BPFP=2.3126 +⌛️ [2/4] FRONTEND: Frontend time: 0.432s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 170, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.847s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 170, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007302 + layer.1.conv_state 0.00049192 0.40243858 + layer.2.ssm_state 0.00000001 0.00000370 + layer.2.conv_state 0.00013328 0.07043453 + layer.3.ssm_state 0.00000001 0.00000505 + layer.3.conv_state 0.00007075 0.06293722 + layer.4.ssm_state 0.00000001 0.00000635 + layer.4.conv_state 0.00027001 0.15136218 + layer.4.output 0.00000204 0.00041328 + ------------------------------------------------------------------------------------- + TOTAL 0.00002499 0.01788958 + (elements=1,515,520) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1515520 +Total Bytes 718152 +BPFP 3.7909 bits/point +EBPFP 6.5193 equivalent bits/point +MSE 0.017890 +---------------------- -------------------------------------------------------- +Time: 1.288s Load: 0.009s, Pack+Encode: 0.432s, Decode+Unpack: 0.847s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0179 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample52-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample52-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample53-layer4-item1.zst (77/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample53-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 200, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 200, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 200, 4096]) -> torch.Size([1, 1, 200, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,528B, BPFP=3.8774 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,284B, BPFP=2.9470 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,200B, BPFP=10.3027 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,116B, BPFP=3.9744 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,372B, BPFP=11.5654 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,648B, BPFP=3.7017 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,092B, BPFP=11.4971 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,468B, BPFP=3.5076 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,828B, BPFP=10.4561 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 223,072B, BPFP=2.1784 +⌛️ [2/4] FRONTEND: Frontend time: 0.446s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 200, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.886s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 200, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000043 0.00007145 + layer.1.conv_state 0.00049042 0.39982879 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00012605 0.06962404 + layer.3.ssm_state 0.00000001 0.00000494 + layer.3.conv_state 0.00011250 0.06192230 + layer.4.ssm_state 0.00000002 0.00000579 + layer.4.conv_state 0.00024020 0.14030373 + layer.4.output 0.00000182 0.00030738 + ------------------------------------------------------------------------------------- + TOTAL 0.00002325 0.01621585 + (elements=1,638,400) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1638400 +Total Bytes 740232 +BPFP 3.6144 bits/point +EBPFP 6.1396 equivalent bits/point +MSE 0.016216 +---------------------- -------------------------------------------------------- +Time: 1.342s Load: 0.010s, Pack+Encode: 0.446s, Decode+Unpack: 0.886s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 200, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0162 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample53-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample53-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample55-layer4-item1.zst (78/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample55-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 199, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 199, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 199, 4096]) -> torch.Size([1, 1, 199, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,496B, BPFP=3.8755 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,288B, BPFP=2.9473 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,156B, BPFP=10.2920 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,824B, BPFP=3.9565 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,588B, BPFP=11.6182 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,784B, BPFP=3.7100 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,296B, BPFP=11.5469 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,000B, BPFP=3.4790 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,948B, BPFP=10.4854 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 230,224B, BPFP=2.2596 +⌛️ [2/4] FRONTEND: Frontend time: 0.463s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 199, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.894s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 199, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000043 0.00007135 + layer.1.conv_state 0.00048601 0.40291613 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00014442 0.07144490 + layer.3.ssm_state 0.00000001 0.00000508 + layer.3.conv_state 0.00011344 0.06341282 + layer.4.ssm_state 0.00000006 0.00000620 + layer.4.conv_state 0.00022676 0.14664069 + layer.4.output 0.00000196 0.00040393 + ------------------------------------------------------------------------------------- + TOTAL 0.00002341 0.01655926 + (elements=1,634,304) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1634304 +Total Bytes 747228 +BPFP 3.6577 bits/point +EBPFP 6.1885 equivalent bits/point +MSE 0.016559 +---------------------- -------------------------------------------------------- +Time: 1.367s Load: 0.010s, Pack+Encode: 0.463s, Decode+Unpack: 0.894s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 199, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0166 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample55-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample55-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample56-layer4-item1.zst (79/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample56-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 215, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 215, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 215, 4096]) -> torch.Size([1, 1, 215, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,520B, BPFP=3.8770 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,136B, BPFP=2.9380 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,156B, BPFP=10.2920 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,736B, BPFP=3.9512 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,676B, BPFP=11.6396 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,392B, BPFP=3.6860 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,256B, BPFP=11.5371 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,912B, BPFP=3.4736 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,980B, BPFP=10.4932 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 241,416B, BPFP=2.1931 +⌛️ [2/4] FRONTEND: Frontend time: 0.449s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 215, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.893s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 215, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000015 0.00007165 + layer.1.conv_state 0.00047793 0.40391606 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00012432 0.07127225 + layer.3.ssm_state 0.00000001 0.00000482 + layer.3.conv_state 0.00007747 0.06405041 + layer.4.ssm_state 0.00000001 0.00000634 + layer.4.conv_state 0.00032616 0.15547413 + layer.4.output 0.00000162 0.00037910 + ------------------------------------------------------------------------------------- + TOTAL 0.00002306 0.01612208 + (elements=1,699,840) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1699840 +Total Bytes 757804 +BPFP 3.5665 bits/point +EBPFP 5.9968 equivalent bits/point +MSE 0.016122 +---------------------- -------------------------------------------------------- +Time: 1.353s Load: 0.011s, Pack+Encode: 0.449s, Decode+Unpack: 0.893s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 215, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0161 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample56-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample56-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample58-layer4-item1.zst (80/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample58-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 174, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 174, 4096]) -> torch.Size([1, 1, 174, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,704B, BPFP=3.8882 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,652B, BPFP=2.9695 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,136B, BPFP=10.2871 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,184B, BPFP=3.9785 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,520B, BPFP=11.6016 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,808B, BPFP=3.7114 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,088B, BPFP=11.4961 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,200B, BPFP=3.4912 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,720B, BPFP=10.4297 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 200,568B, BPFP=2.2513 +⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 174, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.858s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 174, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007340 + layer.1.conv_state 0.00048286 0.40338710 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013493 0.07013075 + layer.3.ssm_state 0.00000001 0.00000495 + layer.3.conv_state 0.00011535 0.06208403 + layer.4.ssm_state 0.00000001 0.00000633 + layer.4.conv_state 0.00022609 0.14507973 + layer.4.output 0.00000231 0.00040153 + ------------------------------------------------------------------------------------- + TOTAL 0.00002474 0.01755838 + (elements=1,531,904) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1531904 +Total Bytes 718204 +BPFP 3.7506 bits/point +EBPFP 6.4539 equivalent bits/point +MSE 0.017558 +---------------------- -------------------------------------------------------- +Time: 1.303s Load: 0.009s, Pack+Encode: 0.435s, Decode+Unpack: 0.858s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0176 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample58-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample58-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample63-layer4-item1.zst (81/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample63-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 186, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 186, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 186, 4096]) -> torch.Size([1, 1, 186, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,512B, BPFP=3.8765 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,232B, BPFP=2.9438 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,200B, BPFP=10.3027 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,856B, BPFP=3.9585 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,512B, BPFP=11.5996 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,756B, BPFP=3.7083 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,252B, BPFP=11.5361 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,196B, BPFP=3.4910 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,052B, BPFP=10.5107 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 205,436B, BPFP=2.1572 +⌛️ [2/4] FRONTEND: Frontend time: 0.443s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 186, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.854s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 186, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007138 + layer.1.conv_state 0.00049416 0.40299881 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00014839 0.07099498 + layer.3.ssm_state 0.00000001 0.00000500 + layer.3.conv_state 0.00011359 0.06316747 + layer.4.ssm_state 0.00000002 0.00000600 + layer.4.conv_state 0.00020754 0.14324577 + layer.4.output 0.00000201 0.00038441 + ------------------------------------------------------------------------------------- + TOTAL 0.00002400 0.01701087 + (elements=1,581,056) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1581056 +Total Bytes 722628 +BPFP 3.6564 bits/point +EBPFP 6.2734 equivalent bits/point +MSE 0.017011 +---------------------- -------------------------------------------------------- +Time: 1.308s Load: 0.010s, Pack+Encode: 0.443s, Decode+Unpack: 0.854s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 186, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0170 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample63-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample63-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample64-layer4-item1.zst (82/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample64-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 185, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 185, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 185, 4096]) -> torch.Size([1, 1, 185, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,664B, BPFP=3.8857 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,360B, BPFP=2.9517 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,112B, BPFP=10.2812 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,168B, BPFP=3.9775 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,440B, BPFP=11.5820 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,132B, BPFP=3.7312 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,032B, BPFP=11.4824 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,836B, BPFP=3.4690 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,700B, BPFP=10.4248 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 204,472B, BPFP=2.1587 +⌛️ [2/4] FRONTEND: Frontend time: 0.443s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 185, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.854s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 185, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000024 0.00007239 + layer.1.conv_state 0.00048011 0.40205610 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00013394 0.06977941 + layer.3.ssm_state 0.00000001 0.00000500 + layer.3.conv_state 0.00011242 0.06190440 + layer.4.ssm_state 0.00000002 0.00000606 + layer.4.conv_state 0.00023441 0.14105755 + layer.4.output 0.00000189 0.00032690 + ------------------------------------------------------------------------------------- + TOTAL 0.00002393 0.01690995 + (elements=1,576,960) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1576960 +Total Bytes 721540 +BPFP 3.6604 bits/point +EBPFP 6.2835 equivalent bits/point +MSE 0.016910 +---------------------- -------------------------------------------------------- +Time: 1.307s Load: 0.009s, Pack+Encode: 0.443s, Decode+Unpack: 0.854s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 185, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0169 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample64-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample64-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample67-layer4-item1.zst (83/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample67-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 179, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 179, 4096]) -> torch.Size([1, 1, 179, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,520B, BPFP=3.8770 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,420B, BPFP=2.9553 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,140B, BPFP=10.2881 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,920B, BPFP=3.9624 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,492B, BPFP=11.5947 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,620B, BPFP=3.7000 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,080B, BPFP=11.4941 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,084B, BPFP=3.4841 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,888B, BPFP=10.4707 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 203,316B, BPFP=2.2184 +⌛️ [2/4] FRONTEND: Frontend time: 0.444s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 179, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.858s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 179, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000030 0.00007318 + layer.1.conv_state 0.00050588 0.40303433 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00011591 0.07055446 + layer.3.ssm_state 0.00000001 0.00000473 + layer.3.conv_state 0.00011742 0.06197603 + layer.4.ssm_state 0.00000003 0.00000606 + layer.4.conv_state 0.00023463 0.14190608 + layer.4.output 0.00000192 0.00041916 + ------------------------------------------------------------------------------------- + TOTAL 0.00002458 0.01727253 + (elements=1,552,384) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1552384 +Total Bytes 720104 +BPFP 3.7110 bits/point +EBPFP 6.3742 equivalent bits/point +MSE 0.017273 +---------------------- -------------------------------------------------------- +Time: 1.311s Load: 0.009s, Pack+Encode: 0.444s, Decode+Unpack: 0.858s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0173 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample67-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample67-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample68-layer4-item1.zst (84/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample68-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 190, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 190, 4096]) -> torch.Size([1, 1, 190, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,536B, BPFP=3.8779 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,284B, BPFP=2.9470 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,160B, BPFP=10.2930 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,132B, BPFP=3.9753 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,492B, BPFP=11.5947 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,472B, BPFP=3.6909 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,144B, BPFP=11.5098 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,556B, BPFP=3.5129 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,968B, BPFP=10.4902 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 200,100B, BPFP=2.0569 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 190, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.865s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 190, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000044 0.00007156 + layer.1.conv_state 0.00050610 0.39909321 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013374 0.06965167 + layer.3.ssm_state 0.00000001 0.00000485 + layer.3.conv_state 0.00010957 0.06222485 + layer.4.ssm_state 0.00000005 0.00000590 + layer.4.conv_state 0.00022355 0.14298116 + layer.4.output 0.00000189 0.00033582 + ------------------------------------------------------------------------------------- + TOTAL 0.00002392 0.01668423 + (elements=1,597,440) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1597440 +Total Bytes 717468 +BPFP 3.5931 bits/point +EBPFP 6.1841 equivalent bits/point +MSE 0.016684 +---------------------- -------------------------------------------------------- +Time: 1.311s Load: 0.012s, Pack+Encode: 0.434s, Decode+Unpack: 0.865s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0167 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample68-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample68-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample69-layer4-item1.zst (85/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample69-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 170, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 170, 4096]) -> torch.Size([1, 1, 170, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,600B, BPFP=3.8818 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,184B, BPFP=2.9409 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,172B, BPFP=10.2959 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,916B, BPFP=3.9622 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,440B, BPFP=11.5820 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,796B, BPFP=3.7107 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,248B, BPFP=11.5352 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,964B, BPFP=3.4768 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,672B, BPFP=10.4180 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 196,076B, BPFP=2.2527 +⌛️ [2/4] FRONTEND: Frontend time: 0.431s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 170, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.854s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 170, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007250 + layer.1.conv_state 0.00047344 0.40493822 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00014015 0.06984502 + layer.3.ssm_state 0.00000001 0.00000488 + layer.3.conv_state 0.00011599 0.06123917 + layer.4.ssm_state 0.00000001 0.00000573 + layer.4.conv_state 0.00022458 0.13775466 + layer.4.output 0.00000213 0.00040964 + ------------------------------------------------------------------------------------- + TOTAL 0.00002480 0.01759816 + (elements=1,515,520) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1515520 +Total Bytes 712692 +BPFP 3.7621 bits/point +EBPFP 6.4892 equivalent bits/point +MSE 0.017598 +---------------------- -------------------------------------------------------- +Time: 1.294s Load: 0.009s, Pack+Encode: 0.431s, Decode+Unpack: 0.854s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 170, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0176 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample69-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample69-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample7-layer4-item1.zst (86/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample7-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 235, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 235, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 235, 4096]) -> torch.Size([1, 1, 235, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,720B, BPFP=3.8892 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,344B, BPFP=2.9507 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,204B, BPFP=10.3037 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,844B, BPFP=3.9578 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,620B, BPFP=11.6260 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,424B, BPFP=3.6880 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,424B, BPFP=11.5781 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,888B, BPFP=3.4722 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,956B, BPFP=10.4873 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 260,556B, BPFP=2.1655 +⌛️ [2/4] FRONTEND: Frontend time: 0.446s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 235, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.887s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 235, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007162 + layer.1.conv_state 0.00050392 0.40041366 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00010721 0.07069305 + layer.3.ssm_state 0.00000001 0.00000502 + layer.3.conv_state 0.00007772 0.06390357 + layer.4.ssm_state 0.00000001 0.00000648 + layer.4.conv_state 0.00021031 0.15365474 + layer.4.output 0.00000154 0.00033038 + ------------------------------------------------------------------------------------- + TOTAL 0.00002006 0.01526075 + (elements=1,781,760) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1781760 +Total Bytes 777604 +BPFP 3.4914 bits/point +EBPFP 5.8129 equivalent bits/point +MSE 0.015261 +---------------------- -------------------------------------------------------- +Time: 1.345s Load: 0.012s, Pack+Encode: 0.446s, Decode+Unpack: 0.887s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 235, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0153 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample7-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample7-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample72-layer4-item1.zst (87/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample72-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 176, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 176, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 176, 4096]) -> torch.Size([1, 1, 176, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,788B, BPFP=3.8933 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,256B, BPFP=2.9453 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,120B, BPFP=10.2832 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,992B, BPFP=3.9668 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,488B, BPFP=11.5938 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,468B, BPFP=3.6907 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 46,916B, BPFP=11.4541 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,552B, BPFP=3.4517 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,616B, BPFP=10.4043 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 198,408B, BPFP=2.2018 +⌛️ [2/4] FRONTEND: Frontend time: 0.442s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 176, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.853s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 176, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007047 + layer.1.conv_state 0.00049605 0.40237013 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00014004 0.06968214 + layer.3.ssm_state 0.00000001 0.00000482 + layer.3.conv_state 0.00011040 0.06109000 + layer.4.ssm_state 0.00000002 0.00000652 + layer.4.conv_state 0.00070183 0.15780047 + layer.4.output 0.00000198 0.00035160 + ------------------------------------------------------------------------------------- + TOTAL 0.00003488 0.01766183 + (elements=1,540,096) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1540096 +Total Bytes 714228 +BPFP 3.7100 bits/point +EBPFP 6.3895 equivalent bits/point +MSE 0.017662 +---------------------- -------------------------------------------------------- +Time: 1.304s Load: 0.009s, Pack+Encode: 0.442s, Decode+Unpack: 0.853s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 176, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0177 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample72-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample72-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample74-layer4-item1.zst (88/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample74-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 172, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 172, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 172, 4096]) -> torch.Size([1, 1, 172, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,644B, BPFP=3.8845 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,188B, BPFP=2.9412 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,132B, BPFP=10.2861 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,988B, BPFP=3.9666 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,532B, BPFP=11.6045 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,624B, BPFP=3.7002 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,180B, BPFP=11.5186 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,600B, BPFP=3.4546 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,836B, BPFP=10.4580 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 199,828B, BPFP=2.2691 +⌛️ [2/4] FRONTEND: Frontend time: 0.444s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 172, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.852s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 172, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007151 + layer.1.conv_state 0.00050574 0.40284139 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00012521 0.07003346 + layer.3.ssm_state 0.00000001 0.00000497 + layer.3.conv_state 0.00011811 0.06254783 + layer.4.ssm_state 0.00000001 0.00000646 + layer.4.conv_state 0.00032558 0.15752517 + layer.4.output 0.00000215 0.00035627 + ------------------------------------------------------------------------------------- + TOTAL 0.00002726 0.01789333 + (elements=1,523,712) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1523712 +Total Bytes 716176 +BPFP 3.7602 bits/point +EBPFP 6.4712 equivalent bits/point +MSE 0.017893 +---------------------- -------------------------------------------------------- +Time: 1.305s Load: 0.009s, Pack+Encode: 0.444s, Decode+Unpack: 0.852s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 172, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0179 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample74-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample74-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample75-layer4-item1.zst (89/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample75-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 164, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 164, 4096]) -> torch.Size([1, 1, 164, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,512B, BPFP=3.8765 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,680B, BPFP=2.9712 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,184B, BPFP=10.2988 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,208B, BPFP=3.9800 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,452B, BPFP=11.5850 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,888B, BPFP=3.7163 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,056B, BPFP=11.4883 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,788B, BPFP=3.4661 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,856B, BPFP=10.4629 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 197,100B, BPFP=2.3473 +⌛️ [2/4] FRONTEND: Frontend time: 0.436s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 164, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.879s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 164, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000255 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007292 + layer.1.conv_state 0.00049170 0.40725982 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00014440 0.07132345 + layer.3.ssm_state 0.00000001 0.00000497 + layer.3.conv_state 0.00007111 0.06341081 + layer.4.ssm_state 0.00000001 0.00000628 + layer.4.conv_state 0.00032601 0.15744065 + layer.4.output 0.00000215 0.00037441 + ------------------------------------------------------------------------------------- + TOTAL 0.00002691 0.01842961 + (elements=1,490,944) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1490944 +Total Bytes 714348 +BPFP 3.8330 bits/point +EBPFP 6.6084 equivalent bits/point +MSE 0.018430 +---------------------- -------------------------------------------------------- +Time: 1.324s Load: 0.009s, Pack+Encode: 0.436s, Decode+Unpack: 0.879s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 164, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0184 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample75-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample75-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample76-layer4-item1.zst (90/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample76-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 166, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) -> torch.Size([1, 1, 166, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,552B, BPFP=3.8789 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,468B, BPFP=2.9583 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,184B, BPFP=10.2988 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,912B, BPFP=3.9619 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,500B, BPFP=11.5967 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,676B, BPFP=3.7034 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,200B, BPFP=11.5234 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,000B, BPFP=3.4790 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,840B, BPFP=10.4590 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 196,364B, BPFP=2.3104 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.858s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 166, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000031 0.00007330 + layer.1.conv_state 0.00048773 0.40322661 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00012531 0.07010893 + layer.3.ssm_state 0.00000001 0.00000491 + layer.3.conv_state 0.00012205 0.06280561 + layer.4.ssm_state 0.00000002 0.00000592 + layer.4.conv_state 0.00022309 0.14145806 + layer.4.output 0.00000212 0.00044419 + ------------------------------------------------------------------------------------- + TOTAL 0.00002513 0.01788531 + (elements=1,499,136) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1499136 +Total Bytes 713320 +BPFP 3.8066 bits/point +EBPFP 6.5653 equivalent bits/point +MSE 0.017885 +---------------------- -------------------------------------------------------- +Time: 1.301s Load: 0.009s, Pack+Encode: 0.434s, Decode+Unpack: 0.858s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 166, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0179 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample76-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample76-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample78-layer4-item1.zst (91/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample78-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 180, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 180, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 180, 4096]) -> torch.Size([1, 1, 180, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,476B, BPFP=3.8743 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,236B, BPFP=2.9441 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,192B, BPFP=10.3008 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,956B, BPFP=3.9646 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,580B, BPFP=11.6162 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 61,040B, BPFP=3.7256 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,256B, BPFP=11.5371 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,256B, BPFP=3.4946 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,096B, BPFP=10.5215 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 211,468B, BPFP=2.2946 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 180, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.856s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 180, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007468 + layer.1.conv_state 0.00049117 0.40583682 + layer.2.ssm_state 0.00000001 0.00000369 + layer.2.conv_state 0.00012889 0.07204966 + layer.3.ssm_state 0.00000001 0.00000502 + layer.3.conv_state 0.00006806 0.06378770 + layer.4.ssm_state 0.00000002 0.00000601 + layer.4.conv_state 0.00022608 0.14131148 + layer.4.output 0.00000211 0.00039067 + ------------------------------------------------------------------------------------- + TOTAL 0.00002335 0.01733093 + (elements=1,556,480) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1556480 +Total Bytes 729180 +BPFP 3.7478 bits/point +EBPFP 6.4088 equivalent bits/point +MSE 0.017331 +---------------------- -------------------------------------------------------- +Time: 1.301s Load: 0.011s, Pack+Encode: 0.434s, Decode+Unpack: 0.856s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 180, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0173 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample78-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample78-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample80-layer4-item1.zst (92/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample80-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 180, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 180, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 180, 4096]) -> torch.Size([1, 1, 180, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,548B, BPFP=3.8787 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,232B, BPFP=2.9438 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,212B, BPFP=10.3057 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,116B, BPFP=3.9744 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,452B, BPFP=11.5850 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,584B, BPFP=3.6978 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,076B, BPFP=11.4932 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,956B, BPFP=3.4763 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,772B, BPFP=10.4424 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 199,220B, BPFP=2.1617 +⌛️ [2/4] FRONTEND: Frontend time: 0.438s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 180, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.854s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 180, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000056 0.00007501 + layer.1.conv_state 0.00050875 0.40648752 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00009667 0.06962049 + layer.3.ssm_state 0.00000001 0.00000491 + layer.3.conv_state 0.00011205 0.06226292 + layer.4.ssm_state 0.00000001 0.00000612 + layer.4.conv_state 0.00021440 0.14538282 + layer.4.output 0.00000203 0.00033521 + ------------------------------------------------------------------------------------- + TOTAL 0.00002371 0.01732086 + (elements=1,556,480) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1556480 +Total Bytes 715792 +BPFP 3.6790 bits/point +EBPFP 6.3341 equivalent bits/point +MSE 0.017321 +---------------------- -------------------------------------------------------- +Time: 1.301s Load: 0.009s, Pack+Encode: 0.438s, Decode+Unpack: 0.854s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 180, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0173 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample80-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample80-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample82-layer4-item1.zst (93/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample82-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 174, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 174, 4096]) -> torch.Size([1, 1, 174, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,420B, BPFP=3.8708 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,072B, BPFP=2.9341 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,088B, BPFP=10.2754 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,868B, BPFP=3.9592 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,464B, BPFP=11.5879 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,720B, BPFP=3.7061 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,344B, BPFP=11.5586 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,608B, BPFP=3.4551 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,184B, BPFP=10.5430 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 201,052B, BPFP=2.2568 +⌛️ [2/4] FRONTEND: Frontend time: 0.433s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 174, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.855s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 174, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007260 + layer.1.conv_state 0.00049114 0.40488496 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00011528 0.07090092 + layer.3.ssm_state 0.00000001 0.00000516 + layer.3.conv_state 0.00007296 0.06373675 + layer.4.ssm_state 0.00000001 0.00000662 + layer.4.conv_state 0.00026103 0.15723972 + layer.4.output 0.00000214 0.00040053 + ------------------------------------------------------------------------------------- + TOTAL 0.00002425 0.01790186 + (elements=1,531,904) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1531904 +Total Bytes 717444 +BPFP 3.7467 bits/point +EBPFP 6.4434 equivalent bits/point +MSE 0.017902 +---------------------- -------------------------------------------------------- +Time: 1.299s Load: 0.011s, Pack+Encode: 0.433s, Decode+Unpack: 0.855s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0179 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample82-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample82-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample85-layer4-item1.zst (94/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample85-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 180, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 180, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 180, 4096]) -> torch.Size([1, 1, 180, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,420B, BPFP=3.8708 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,156B, BPFP=2.9392 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,180B, BPFP=10.2979 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,456B, BPFP=3.9341 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,724B, BPFP=11.6514 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,404B, BPFP=3.6868 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,536B, BPFP=11.6055 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,660B, BPFP=3.5193 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,980B, BPFP=10.4932 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 205,524B, BPFP=2.2301 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 180, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.853s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 180, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000254 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000037 0.00007167 + layer.1.conv_state 0.00049028 0.40250811 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00010997 0.07133538 + layer.3.ssm_state 0.00000001 0.00000492 + layer.3.conv_state 0.00007961 0.06341118 + layer.4.ssm_state 0.00000002 0.00000566 + layer.4.conv_state 0.00024454 0.13499129 + layer.4.output 0.00000217 0.00035803 + ------------------------------------------------------------------------------------- + TOTAL 0.00002360 0.01708908 + (elements=1,556,480) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1556480 +Total Bytes 722664 +BPFP 3.7144 bits/point +EBPFP 6.3723 equivalent bits/point +MSE 0.017089 +---------------------- -------------------------------------------------------- +Time: 1.296s Load: 0.009s, Pack+Encode: 0.434s, Decode+Unpack: 0.853s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 180, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0171 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample85-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample85-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample87-layer4-item1.zst (95/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample87-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 179, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 179, 4096]) -> torch.Size([1, 1, 179, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,492B, BPFP=3.8752 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,040B, BPFP=2.9321 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,008B, BPFP=10.2559 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,792B, BPFP=3.9546 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,500B, BPFP=11.5967 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,700B, BPFP=3.7048 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,216B, BPFP=11.5273 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,464B, BPFP=3.4463 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,856B, BPFP=10.4629 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 201,708B, BPFP=2.2009 +⌛️ [2/4] FRONTEND: Frontend time: 0.433s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 179, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.862s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 179, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007129 + layer.1.conv_state 0.00049519 0.40370700 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00011503 0.07082102 + layer.3.ssm_state 0.00000001 0.00000504 + layer.3.conv_state 0.00006908 0.06311605 + layer.4.ssm_state 0.00000001 0.00000637 + layer.4.conv_state 0.00022394 0.15158398 + layer.4.output 0.00000194 0.00041758 + ------------------------------------------------------------------------------------- + TOTAL 0.00002309 0.01751985 + (elements=1,552,384) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1552384 +Total Bytes 717400 +BPFP 3.6970 bits/point +EBPFP 6.3546 equivalent bits/point +MSE 0.017520 +---------------------- -------------------------------------------------------- +Time: 1.305s Load: 0.010s, Pack+Encode: 0.433s, Decode+Unpack: 0.862s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 179, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0175 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample87-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample87-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample9-layer4-item1.zst (96/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample9-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 205, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.012s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 205, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 205, 4096]) -> torch.Size([1, 1, 205, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,460B, BPFP=3.8733 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,608B, BPFP=2.9668 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,176B, BPFP=10.2969 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,948B, BPFP=3.9641 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,512B, BPFP=11.5996 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,432B, BPFP=3.6885 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,128B, BPFP=11.5059 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,024B, BPFP=3.4805 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,784B, BPFP=10.4453 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 237,008B, BPFP=2.2581 +⌛️ [2/4] FRONTEND: Frontend time: 0.450s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 205, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.905s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 205, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007286 + layer.1.conv_state 0.00050539 0.40422124 + layer.2.ssm_state 0.00000001 0.00000369 + layer.2.conv_state 0.00015098 0.07163182 + layer.3.ssm_state 0.00000001 0.00000480 + layer.3.conv_state 0.00007050 0.06318675 + layer.4.ssm_state 0.00000001 0.00000613 + layer.4.conv_state 0.00024328 0.14968339 + layer.4.output 0.00000184 0.00034050 + ------------------------------------------------------------------------------------- + TOTAL 0.00002299 0.01637303 + (elements=1,658,880) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1658880 +Total Bytes 753704 +BPFP 3.6348 bits/point +EBPFP 6.1265 equivalent bits/point +MSE 0.016373 +---------------------- -------------------------------------------------------- +Time: 1.367s Load: 0.012s, Pack+Encode: 0.450s, Decode+Unpack: 0.905s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 205, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0164 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample9-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample9-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample93-layer4-item1.zst (97/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample93-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 174, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 174, 4096]) -> torch.Size([1, 1, 174, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,600B, BPFP=3.8818 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,340B, BPFP=2.9504 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,116B, BPFP=10.2822 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,888B, BPFP=3.9604 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,528B, BPFP=11.6035 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,632B, BPFP=3.7007 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,172B, BPFP=11.5166 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,672B, BPFP=3.4590 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,804B, BPFP=10.4502 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 201,296B, BPFP=2.2595 +⌛️ [2/4] FRONTEND: Frontend time: 0.436s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 174, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.863s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 174, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000253 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007159 + layer.1.conv_state 0.00049634 0.40453181 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00012038 0.07086695 + layer.3.ssm_state 0.00000001 0.00000514 + layer.3.conv_state 0.00007294 0.06338233 + layer.4.ssm_state 0.00000001 0.00000622 + layer.4.conv_state 0.00026211 0.15283602 + layer.4.output 0.00000210 0.00040610 + ------------------------------------------------------------------------------------- + TOTAL 0.00002447 0.01779427 + (elements=1,531,904) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1531904 +Total Bytes 717672 +BPFP 3.7479 bits/point +EBPFP 6.4445 equivalent bits/point +MSE 0.017794 +---------------------- -------------------------------------------------------- +Time: 1.308s Load: 0.009s, Pack+Encode: 0.436s, Decode+Unpack: 0.863s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 174, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0178 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample93-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample93-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample94-layer4-item1.zst (98/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample94-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 190, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 190, 4096]) -> torch.Size([1, 1, 190, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,508B, BPFP=3.8762 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,144B, BPFP=2.9385 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,048B, BPFP=10.2656 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,940B, BPFP=3.9636 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,384B, BPFP=11.5684 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,752B, BPFP=3.7080 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 46,972B, BPFP=11.4678 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,616B, BPFP=3.5166 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,604B, BPFP=10.4014 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 202,308B, BPFP=2.0796 +⌛️ [2/4] FRONTEND: Frontend time: 0.440s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 190, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.857s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 190, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007373 + layer.1.conv_state 0.00050538 0.40369874 + layer.2.ssm_state 0.00000001 0.00000357 + layer.2.conv_state 0.00014003 0.06980760 + layer.3.ssm_state 0.00000001 0.00000487 + layer.3.conv_state 0.00010410 0.06031869 + layer.4.ssm_state 0.00000002 0.00000564 + layer.4.conv_state 0.00021663 0.13511188 + layer.4.output 0.00000194 0.00034342 + ------------------------------------------------------------------------------------- + TOTAL 0.00002379 0.01658524 + (elements=1,597,440) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1597440 +Total Bytes 718900 +BPFP 3.6003 bits/point +EBPFP 6.1874 equivalent bits/point +MSE 0.016585 +---------------------- -------------------------------------------------------- +Time: 1.307s Load: 0.010s, Pack+Encode: 0.440s, Decode+Unpack: 0.857s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 190, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0166 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample94-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample94-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample95-layer4-item1.zst (99/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample95-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 183, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 183, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) -> torch.Size([1, 1, 183, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,428B, BPFP=3.8713 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,204B, BPFP=2.9421 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,140B, BPFP=10.2881 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,752B, BPFP=3.9521 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,628B, BPFP=11.6279 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,760B, BPFP=3.7085 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,140B, BPFP=11.5088 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,744B, BPFP=3.4634 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,772B, BPFP=10.4424 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 199,904B, BPFP=2.1335 +⌛️ [2/4] FRONTEND: Frontend time: 0.441s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.854s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 183, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000041 0.00007041 + layer.1.conv_state 0.00050614 0.40283760 + layer.2.ssm_state 0.00000001 0.00000366 + layer.2.conv_state 0.00012733 0.07098902 + layer.3.ssm_state 0.00000001 0.00000503 + layer.3.conv_state 0.00007038 0.06280443 + layer.4.ssm_state 0.00000001 0.00000644 + layer.4.conv_state 0.00022687 0.14626490 + layer.4.output 0.00000194 0.00040089 + ------------------------------------------------------------------------------------- + TOTAL 0.00002346 0.01720093 + (elements=1,568,768) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1568768 +Total Bytes 716096 +BPFP 3.6518 bits/point +EBPFP 6.2841 equivalent bits/point +MSE 0.017201 +---------------------- -------------------------------------------------------- +Time: 1.306s Load: 0.011s, Pack+Encode: 0.441s, Decode+Unpack: 0.854s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 183, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0172 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample95-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample95-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample99-layer4-item1.zst (100/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample99-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.011s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 191, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 191, 4096]) -> torch.Size([1, 1, 191, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,576B, BPFP=3.8804 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,260B, BPFP=2.9456 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,148B, BPFP=10.2900 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,092B, BPFP=3.9729 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,576B, BPFP=11.6152 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,484B, BPFP=3.6917 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,208B, BPFP=11.5254 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,440B, BPFP=3.5059 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,868B, BPFP=10.4658 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 197,168B, BPFP=2.0162 +⌛️ [2/4] FRONTEND: Frontend time: 0.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 191, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.864s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 191, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000040 0.00007190 + layer.1.conv_state 0.00049989 0.39859003 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00013469 0.06989554 + layer.3.ssm_state 0.00000001 0.00000492 + layer.3.conv_state 0.00011145 0.06181328 + layer.4.ssm_state 0.00000005 0.00000579 + layer.4.conv_state 0.00022056 0.14219369 + layer.4.output 0.00000184 0.00032315 + ------------------------------------------------------------------------------------- + TOTAL 0.00002370 0.01660642 + (elements=1,601,536) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1601536 +Total Bytes 714444 +BPFP 3.5688 bits/point +EBPFP 6.1527 equivalent bits/point +MSE 0.016606 +---------------------- -------------------------------------------------------- +Time: 1.309s Load: 0.011s, Pack+Encode: 0.434s, Decode+Unpack: 0.864s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 191, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0166 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_truthfulqa_mc1/sample99-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample99-layer4-item1.zst +------------------------ ---------------------------- +TOTAL PROCESSING SUMMARY +------------------------ ---------------------------- +Total files 100 +Avg BPFP 3.6799 bits/point +Avg EBPFP 6.2933 equivalent bits/point +Avg MSE 0.017113 +Avg Time 1.337s +------------------------ ---------------------------- diff --git a/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/dtufc_hyperprior-featurecoding_falconmamba_individual.log b/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/dtufc_hyperprior-featurecoding_falconmamba_individual.log new file mode 100644 index 0000000000000000000000000000000000000000..b192537e9f896a8087b9ad868d3cb47f321f24b1 --- /dev/null +++ b/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/dtufc_hyperprior-featurecoding_falconmamba_individual.log @@ -0,0 +1,16858 @@ +Experiment: dtufc_hyperprior-featurecoding_falconmamba_individual +Log file: output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/dtufc_hyperprior-featurecoding_falconmamba_individual.log +DTUFCCodecConfig: + arch: hyperprior-featurecoding + handler: falconmamba + checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.02_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar + transform_type: kmeans + transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json + bit_depth: 8 + device: cuda:0 +Loading checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.02_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Checkpoint epoch: 598 +Loaded hyperprior-featurecoding (1-channel) on cuda:0 +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.0.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.0.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_0_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.1.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.1.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_1_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.2.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.2.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_2_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.3.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.3.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_3_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_conv.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.4.conv_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_conv.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_ssm.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.4.ssm_state' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_layer_4_ssm.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_output.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.4.output' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/falconmamba/arc_fewshot-8bit_output.json +Loaded per-key mappings: model=falconmamba + Keys: ['layer.0.conv_state', 'layer.0.ssm_state', 'layer.1.conv_state', 'layer.1.ssm_state', 'layer.2.conv_state', 'layer.2.ssm_state', 'layer.3.conv_state', 'layer.3.ssm_state', 'layer.4.conv_state', 'layer.4.ssm_state', 'layer.4.output'] +---------------- ----------------------------------------------------------------------------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.02_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Transform type kmeans +Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json +Input ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande +Output output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande +---------------- ----------------------------------------------------------------------------------------------------------------------------- +Files found: 100 +---------------------------------------------------------------------- + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample1-layer4-item1.zst (1/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample1-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 109, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 109, 4096]) -> torch.Size([1, 1, 109, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,576B, BPFP=3.8804 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,640B, BPFP=2.9688 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,212B, BPFP=10.3057 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,144B, BPFP=3.9761 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,568B, BPFP=11.6133 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,672B, BPFP=3.7031 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,176B, BPFP=11.5176 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,260B, BPFP=3.4949 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,920B, BPFP=10.4785 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 134,404B, BPFP=2.4083 +⌛️ [2/4] FRONTEND: Frontend time: 0.650s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 109, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.850s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 109, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007315 + layer.1.conv_state 0.00050886 0.40198040 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013788 0.07075954 + layer.3.ssm_state 0.00000001 0.00000469 + layer.3.conv_state 0.00007003 0.06338074 + layer.4.ssm_state 0.00000003 0.00000621 + layer.4.conv_state 0.00023930 0.13966891 + layer.4.output 0.00000273 0.00049399 + ------------------------------------------------------------------------------------- + TOTAL 0.00002952 0.02107331 + (elements=1,265,664) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1265664 +Total Bytes 652196 +BPFP 4.1224 bits/point +EBPFP 7.3953 equivalent bits/point +MSE 0.021073 +---------------------- -------------------------------------------------------- +Time: 1.506s Load: 0.007s, Pack+Encode: 0.650s, Decode+Unpack: 0.850s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0211 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample1-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample1-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample10-layer4-item1.zst (2/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample10-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 105, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,496B, BPFP=3.8755 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,456B, BPFP=2.9575 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,216B, BPFP=10.3066 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,128B, BPFP=3.9751 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,640B, BPFP=11.6309 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,508B, BPFP=3.6931 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,220B, BPFP=11.5283 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,332B, BPFP=3.4993 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,904B, BPFP=10.4746 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 131,680B, BPFP=2.4494 +⌛️ [2/4] FRONTEND: Frontend time: 0.428s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.824s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007302 + layer.1.conv_state 0.00050781 0.40232232 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013544 0.07119112 + layer.3.ssm_state 0.00000001 0.00000466 + layer.3.conv_state 0.00006839 0.06270153 + layer.4.ssm_state 0.00000001 0.00000585 + layer.4.conv_state 0.00024756 0.13537189 + layer.4.output 0.00000285 0.00051890 + ------------------------------------------------------------------------------------- + TOTAL 0.00002999 0.02124149 + (elements=1,249,280) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1249280 +Total Bytes 649204 +BPFP 4.1573 bits/point +EBPFP 7.4714 equivalent bits/point +MSE 0.021241 +---------------------- -------------------------------------------------------- +Time: 1.259s Load: 0.008s, Pack+Encode: 0.428s, Decode+Unpack: 0.824s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0212 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample10-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample10-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample100-layer4-item1.zst (3/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample100-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 98, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,440B, BPFP=3.8721 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,568B, BPFP=2.9644 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,256B, BPFP=10.3164 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,044B, BPFP=3.9700 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,576B, BPFP=11.6152 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,748B, BPFP=3.7078 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,152B, BPFP=11.5117 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,456B, BPFP=3.5068 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,944B, BPFP=10.4844 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,136B, BPFP=2.4541 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.824s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000014 0.00007362 + layer.1.conv_state 0.00050623 0.40375978 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013817 0.07133129 + layer.3.ssm_state 0.00000001 0.00000480 + layer.3.conv_state 0.00007075 0.06303844 + layer.4.ssm_state 0.00000001 0.00000577 + layer.4.conv_state 0.00027926 0.13317038 + layer.4.output 0.00000298 0.00066384 + ------------------------------------------------------------------------------------- + TOTAL 0.00003162 0.02176830 + (elements=1,220,608) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1220608 +Total Bytes 640944 +BPFP 4.2008 bits/point +EBPFP 7.5946 equivalent bits/point +MSE 0.021768 +---------------------- -------------------------------------------------------- +Time: 1.251s Load: 0.008s, Pack+Encode: 0.419s, Decode+Unpack: 0.824s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample100-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample100-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample103-layer4-item1.zst (4/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample103-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,520B, BPFP=3.8770 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,616B, BPFP=2.9673 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,188B, BPFP=10.2998 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,240B, BPFP=3.9819 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,592B, BPFP=11.6191 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,600B, BPFP=3.6987 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,168B, BPFP=11.5156 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,432B, BPFP=3.5054 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,920B, BPFP=10.4785 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,372B, BPFP=2.4581 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.840s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007415 + layer.1.conv_state 0.00051199 0.40324616 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013382 0.07138020 + layer.3.ssm_state 0.00000001 0.00000462 + layer.3.conv_state 0.00007055 0.06292976 + layer.4.ssm_state 0.00000005 0.00000597 + layer.4.conv_state 0.00025245 0.13858618 + layer.4.output 0.00000298 0.00065333 + ------------------------------------------------------------------------------------- + TOTAL 0.00003058 0.02161355 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 646272 +BPFP 4.1796 bits/point +EBPFP 7.5291 equivalent bits/point +MSE 0.021614 +---------------------- -------------------------------------------------------- +Time: 1.266s Load: 0.007s, Pack+Encode: 0.419s, Decode+Unpack: 0.840s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0216 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample103-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample103-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample104-layer4-item1.zst (5/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample104-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,520B, BPFP=3.8770 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,608B, BPFP=2.9668 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,148B, BPFP=10.2900 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,060B, BPFP=3.9709 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,520B, BPFP=11.6016 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,540B, BPFP=3.6951 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,016B, BPFP=11.4785 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,352B, BPFP=3.5005 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,768B, BPFP=10.4414 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,516B, BPFP=2.4370 +⌛️ [2/4] FRONTEND: Frontend time: 0.437s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.823s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007350 + layer.1.conv_state 0.00050971 0.40150660 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00014652 0.07057760 + layer.3.ssm_state 0.00000001 0.00000471 + layer.3.conv_state 0.00006897 0.06309742 + layer.4.ssm_state 0.00000002 0.00000597 + layer.4.conv_state 0.00023369 0.13668135 + layer.4.output 0.00000300 0.00065136 + ------------------------------------------------------------------------------------- + TOTAL 0.00003021 0.02143066 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 645672 +BPFP 4.1620 bits/point +EBPFP 7.4955 equivalent bits/point +MSE 0.021431 +---------------------- -------------------------------------------------------- +Time: 1.267s Load: 0.007s, Pack+Encode: 0.437s, Decode+Unpack: 0.823s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample104-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample104-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample105-layer4-item1.zst (6/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample105-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 101, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,616B, BPFP=3.8828 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,500B, BPFP=2.9602 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,152B, BPFP=10.2910 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,184B, BPFP=3.9785 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,460B, BPFP=11.5869 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,616B, BPFP=3.6997 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,064B, BPFP=11.4902 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,256B, BPFP=3.4946 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,832B, BPFP=10.4570 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,316B, BPFP=2.4233 +⌛️ [2/4] FRONTEND: Frontend time: 0.421s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007317 + layer.1.conv_state 0.00050648 0.40141001 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00011943 0.07040541 + layer.3.ssm_state 0.00000001 0.00000470 + layer.3.conv_state 0.00007205 0.06244205 + layer.4.ssm_state 0.00000003 0.00000617 + layer.4.conv_state 0.00022891 0.13859114 + layer.4.output 0.00000312 0.00052268 + ------------------------------------------------------------------------------------- + TOTAL 0.00002958 0.02155174 + (elements=1,232,896) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1232896 +Total Bytes 642620 +BPFP 4.1698 bits/point +EBPFP 7.5265 equivalent bits/point +MSE 0.021552 +---------------------- -------------------------------------------------------- +Time: 1.250s Load: 0.007s, Pack+Encode: 0.421s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0216 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample105-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample105-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample108-layer4-item1.zst (7/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample108-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 99, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,540B, BPFP=3.8782 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,544B, BPFP=2.9629 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,236B, BPFP=10.3115 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,080B, BPFP=3.9722 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,520B, BPFP=11.6016 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,628B, BPFP=3.7004 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,172B, BPFP=11.5166 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,632B, BPFP=3.5176 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,008B, BPFP=10.5000 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 126,760B, BPFP=2.5008 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000252 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007321 + layer.1.conv_state 0.00050697 0.40168864 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00013442 0.07090984 + layer.3.ssm_state 0.00000001 0.00000474 + layer.3.conv_state 0.00006950 0.06276967 + layer.4.ssm_state 0.00000001 0.00000591 + layer.4.conv_state 0.00023016 0.13333768 + layer.4.output 0.00000331 0.00066685 + ------------------------------------------------------------------------------------- + TOTAL 0.00003020 0.02162927 + (elements=1,224,704) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1224704 +Total Bytes 644744 +BPFP 4.2116 bits/point +EBPFP 7.5952 equivalent bits/point +MSE 0.021629 +---------------------- -------------------------------------------------------- +Time: 1.249s Load: 0.008s, Pack+Encode: 0.420s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0216 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample108-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample108-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample110-layer4-item1.zst (8/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample110-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,540B, BPFP=3.8782 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,552B, BPFP=2.9634 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,184B, BPFP=10.2988 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,060B, BPFP=3.9709 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,668B, BPFP=11.6377 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,672B, BPFP=3.7031 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,148B, BPFP=11.5107 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,416B, BPFP=3.5044 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,928B, BPFP=10.4805 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,344B, BPFP=2.4337 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007415 + layer.1.conv_state 0.00051140 0.40325829 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00014293 0.07178769 + layer.3.ssm_state 0.00000001 0.00000478 + layer.3.conv_state 0.00006973 0.06361722 + layer.4.ssm_state 0.00000001 0.00000604 + layer.4.conv_state 0.00023680 0.13547370 + layer.4.output 0.00000289 0.00065357 + ------------------------------------------------------------------------------------- + TOTAL 0.00003023 0.02149154 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 646136 +BPFP 4.1650 bits/point +EBPFP 7.5026 equivalent bits/point +MSE 0.021492 +---------------------- -------------------------------------------------------- +Time: 1.249s Load: 0.008s, Pack+Encode: 0.418s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample110-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample110-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample111-layer4-item1.zst (9/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample111-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,612B, BPFP=3.8826 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,536B, BPFP=2.9624 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,224B, BPFP=10.3086 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,100B, BPFP=3.9734 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,568B, BPFP=11.6133 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,832B, BPFP=3.7129 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,180B, BPFP=11.5186 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,308B, BPFP=3.4978 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,920B, BPFP=10.4785 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,536B, BPFP=2.4373 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000014 0.00007421 + layer.1.conv_state 0.00051350 0.40336826 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00014131 0.07173721 + layer.3.ssm_state 0.00000001 0.00000480 + layer.3.conv_state 0.00006915 0.06356606 + layer.4.ssm_state 0.00000002 0.00000613 + layer.4.conv_state 0.00025593 0.13735147 + layer.4.output 0.00000302 0.00066357 + ------------------------------------------------------------------------------------- + TOTAL 0.00003077 0.02154475 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 646440 +BPFP 4.1669 bits/point +EBPFP 7.5053 equivalent bits/point +MSE 0.021545 +---------------------- -------------------------------------------------------- +Time: 1.247s Load: 0.008s, Pack+Encode: 0.419s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample111-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample111-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample112-layer4-item1.zst (10/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample112-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,500B, BPFP=3.8757 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,540B, BPFP=2.9626 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,252B, BPFP=10.3154 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,984B, BPFP=3.9663 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,560B, BPFP=11.6113 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,716B, BPFP=3.7058 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,108B, BPFP=11.5010 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,576B, BPFP=3.5142 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,864B, BPFP=10.4648 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 121,044B, BPFP=2.4373 +⌛️ [2/4] FRONTEND: Frontend time: 0.415s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000015 0.00007380 + layer.1.conv_state 0.00050852 0.40174443 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013698 0.07136884 + layer.3.ssm_state 0.00000001 0.00000463 + layer.3.conv_state 0.00006971 0.06260876 + layer.4.ssm_state 0.00000001 0.00000570 + layer.4.conv_state 0.00027123 0.13345172 + layer.4.output 0.00000318 0.00057995 + ------------------------------------------------------------------------------------- + TOTAL 0.00003157 0.02175468 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 638768 +BPFP 4.2007 bits/point +EBPFP 7.6053 equivalent bits/point +MSE 0.021755 +---------------------- -------------------------------------------------------- +Time: 1.241s Load: 0.008s, Pack+Encode: 0.415s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample112-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample112-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample113-layer4-item1.zst (11/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample113-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,544B, BPFP=3.8784 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,624B, BPFP=2.9678 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,260B, BPFP=10.3174 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,056B, BPFP=3.9707 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,516B, BPFP=11.6006 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,600B, BPFP=3.6987 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,152B, BPFP=11.5117 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,856B, BPFP=3.4702 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,820B, BPFP=10.4541 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 118,580B, BPFP=2.3876 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007286 + layer.1.conv_state 0.00049980 0.40330225 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00011977 0.07139204 + layer.3.ssm_state 0.00000001 0.00000487 + layer.3.conv_state 0.00007203 0.06309193 + layer.4.ssm_state 0.00000001 0.00000622 + layer.4.conv_state 0.00024562 0.14186172 + layer.4.output 0.00000308 0.00056081 + ------------------------------------------------------------------------------------- + TOTAL 0.00003020 0.02203054 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 635632 +BPFP 4.1800 bits/point +EBPFP 7.5803 equivalent bits/point +MSE 0.022031 +---------------------- -------------------------------------------------------- +Time: 1.246s Load: 0.008s, Pack+Encode: 0.417s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0220 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample113-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample113-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample117-layer4-item1.zst (12/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample117-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.006s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 95, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,472B, BPFP=3.8740 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,560B, BPFP=2.9639 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,272B, BPFP=10.3203 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,072B, BPFP=3.9717 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,536B, BPFP=11.6055 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,600B, BPFP=3.6987 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,100B, BPFP=11.4990 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,836B, BPFP=3.5300 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,800B, BPFP=10.4492 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 121,988B, BPFP=2.5080 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 95, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.816s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 95, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007358 + layer.1.conv_state 0.00051224 0.40189797 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013107 0.07136193 + layer.3.ssm_state 0.00000001 0.00000472 + layer.3.conv_state 0.00006945 0.06289668 + layer.4.ssm_state 0.00000001 0.00000599 + layer.4.conv_state 0.00027531 0.13145038 + layer.4.output 0.00000313 0.00063278 + ------------------------------------------------------------------------------------- + TOTAL 0.00003178 0.02187278 + (elements=1,208,320) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1208320 +Total Bytes 639860 +BPFP 4.2364 bits/point +EBPFP 7.6651 equivalent bits/point +MSE 0.021873 +---------------------- -------------------------------------------------------- +Time: 1.239s Load: 0.006s, Pack+Encode: 0.417s, Decode+Unpack: 0.816s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0219 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample117-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample117-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample121-layer4-item1.zst (13/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample121-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 94, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 94, 4096]) -> torch.Size([1, 1, 94, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,464B, BPFP=3.8735 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,532B, BPFP=2.9622 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,220B, BPFP=10.3076 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,008B, BPFP=3.9678 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,552B, BPFP=11.6094 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,512B, BPFP=3.6934 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,060B, BPFP=11.4893 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,432B, BPFP=3.5054 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,944B, BPFP=10.4844 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 118,980B, BPFP=2.4722 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 94, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.831s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 94, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000009 0.00007189 + layer.1.conv_state 0.00049706 0.40211016 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00015153 0.07105638 + layer.3.ssm_state 0.00000001 0.00000459 + layer.3.conv_state 0.00007501 0.06243692 + layer.4.ssm_state 0.00000002 0.00000581 + layer.4.conv_state 0.00026177 0.13616519 + layer.4.output 0.00000322 0.00064901 + ------------------------------------------------------------------------------------- + TOTAL 0.00003183 0.02206324 + (elements=1,204,224) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1204224 +Total Bytes 636328 +BPFP 4.2273 bits/point +EBPFP 7.6642 equivalent bits/point +MSE 0.022063 +---------------------- -------------------------------------------------------- +Time: 1.256s Load: 0.007s, Pack+Encode: 0.418s, Decode+Unpack: 0.831s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0221 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample121-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample121-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample123-layer4-item1.zst (14/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample123-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,320B, BPFP=3.8647 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,700B, BPFP=2.9724 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,188B, BPFP=10.2998 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,008B, BPFP=3.9678 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,504B, BPFP=11.5977 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,480B, BPFP=3.6914 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,336B, BPFP=11.5566 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,608B, BPFP=3.5161 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,076B, BPFP=10.5166 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 126,816B, BPFP=2.4769 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.814s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007469 + layer.1.conv_state 0.00051295 0.40159845 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00015130 0.07110754 + layer.3.ssm_state 0.00000001 0.00000470 + layer.3.conv_state 0.00007484 0.06293952 + layer.4.ssm_state 0.00000001 0.00000588 + layer.4.conv_state 0.00021889 0.13400888 + layer.4.output 0.00000110 0.00053903 + ------------------------------------------------------------------------------------- + TOTAL 0.00002983 0.02154223 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 644660 +BPFP 4.1970 bits/point +EBPFP 7.5684 equivalent bits/point +MSE 0.021542 +---------------------- -------------------------------------------------------- +Time: 1.241s Load: 0.007s, Pack+Encode: 0.420s, Decode+Unpack: 0.814s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample123-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample123-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample124-layer4-item1.zst (15/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample124-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 98, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,472B, BPFP=3.8740 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,648B, BPFP=2.9692 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,252B, BPFP=10.3154 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,052B, BPFP=3.9705 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,584B, BPFP=11.6172 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,408B, BPFP=3.6870 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,236B, BPFP=11.5322 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,452B, BPFP=3.5066 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,776B, BPFP=10.4434 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 122,936B, BPFP=2.4501 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.831s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007338 + layer.1.conv_state 0.00051116 0.40241969 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00011994 0.07094650 + layer.3.ssm_state 0.00000001 0.00000470 + layer.3.conv_state 0.00007252 0.06284653 + layer.4.ssm_state 0.00000001 0.00000595 + layer.4.conv_state 0.00026041 0.13490863 + layer.4.output 0.00000302 0.00065789 + ------------------------------------------------------------------------------------- + TOTAL 0.00003081 0.02176153 + (elements=1,220,608) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1220608 +Total Bytes 640440 +BPFP 4.1975 bits/point +EBPFP 7.5893 equivalent bits/point +MSE 0.021762 +---------------------- -------------------------------------------------------- +Time: 1.256s Load: 0.008s, Pack+Encode: 0.417s, Decode+Unpack: 0.831s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample124-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample124-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample125-layer4-item1.zst (16/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample125-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,576B, BPFP=3.8804 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,740B, BPFP=2.9749 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,248B, BPFP=10.3145 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,032B, BPFP=3.9692 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,584B, BPFP=11.6172 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,756B, BPFP=3.7083 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,088B, BPFP=11.4961 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,628B, BPFP=3.5173 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,840B, BPFP=10.4590 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,604B, BPFP=2.4141 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.875s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007301 + layer.1.conv_state 0.00051289 0.40369707 + layer.2.ssm_state 0.00000001 0.00000358 + layer.2.conv_state 0.00013364 0.07109377 + layer.3.ssm_state 0.00000001 0.00000468 + layer.3.conv_state 0.00007097 0.06266086 + layer.4.ssm_state 0.00000004 0.00000585 + layer.4.conv_state 0.00024017 0.13316900 + layer.4.output 0.00000292 0.00053008 + ------------------------------------------------------------------------------------- + TOTAL 0.00003044 0.02156483 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 641720 +BPFP 4.1779 bits/point +EBPFP 7.5510 equivalent bits/point +MSE 0.021565 +---------------------- -------------------------------------------------------- +Time: 1.302s Load: 0.007s, Pack+Encode: 0.420s, Decode+Unpack: 0.875s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0216 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample125-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample125-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample126-layer4-item1.zst (17/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample126-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 98, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,396B, BPFP=3.8694 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,652B, BPFP=2.9695 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,228B, BPFP=10.3096 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,996B, BPFP=3.9670 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,540B, BPFP=11.6064 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,540B, BPFP=3.6951 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,216B, BPFP=11.5273 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,328B, BPFP=3.4990 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,884B, BPFP=10.4697 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,472B, BPFP=2.4608 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.821s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007319 + layer.1.conv_state 0.00050368 0.40310290 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00013069 0.07107789 + layer.3.ssm_state 0.00000001 0.00000466 + layer.3.conv_state 0.00007150 0.06309240 + layer.4.ssm_state 0.00000001 0.00000592 + layer.4.conv_state 0.00023706 0.13460328 + layer.4.output 0.00000305 0.00066582 + ------------------------------------------------------------------------------------- + TOTAL 0.00003027 0.02178438 + (elements=1,220,608) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1220608 +Total Bytes 640876 +BPFP 4.2004 bits/point +EBPFP 7.5915 equivalent bits/point +MSE 0.021784 +---------------------- -------------------------------------------------------- +Time: 1.248s Load: 0.008s, Pack+Encode: 0.420s, Decode+Unpack: 0.821s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample126-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample126-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample13-layer4-item1.zst (18/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample13-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,460B, BPFP=3.8733 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,708B, BPFP=2.9729 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,040B, BPFP=3.9697 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,492B, BPFP=11.5947 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,736B, BPFP=3.7070 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,124B, BPFP=11.5049 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,276B, BPFP=3.4958 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,020B, BPFP=10.5029 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 126,780B, BPFP=2.4041 +⌛️ [2/4] FRONTEND: Frontend time: 0.422s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.821s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000246 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007199 + layer.1.conv_state 0.00051308 0.40156120 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00013634 0.07064953 + layer.3.ssm_state 0.00000001 0.00000470 + layer.3.conv_state 0.00007240 0.06312805 + layer.4.ssm_state 0.00000002 0.00000598 + layer.4.conv_state 0.00022022 0.13476419 + layer.4.output 0.00000307 0.00065123 + ------------------------------------------------------------------------------------- + TOTAL 0.00002979 0.02138399 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 644468 +BPFP 4.1542 bits/point +EBPFP 7.4912 equivalent bits/point +MSE 0.021384 +---------------------- -------------------------------------------------------- +Time: 1.251s Load: 0.007s, Pack+Encode: 0.422s, Decode+Unpack: 0.821s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample13-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample13-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample131-layer4-item1.zst (19/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample131-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,588B, BPFP=3.8811 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,576B, BPFP=2.9648 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,216B, BPFP=10.3066 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,072B, BPFP=3.9717 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,492B, BPFP=11.5947 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,596B, BPFP=3.6985 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,144B, BPFP=11.5098 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,952B, BPFP=3.4761 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,832B, BPFP=10.4570 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 118,652B, BPFP=2.3891 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007240 + layer.1.conv_state 0.00049676 0.40469515 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00011509 0.07127514 + layer.3.ssm_state 0.00000001 0.00000479 + layer.3.conv_state 0.00006928 0.06293685 + layer.4.ssm_state 0.00000002 0.00000619 + layer.4.conv_state 0.00024590 0.14002413 + layer.4.output 0.00000308 0.00056423 + ------------------------------------------------------------------------------------- + TOTAL 0.00002993 0.02201229 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 635744 +BPFP 4.1808 bits/point +EBPFP 7.5813 equivalent bits/point +MSE 0.022012 +---------------------- -------------------------------------------------------- +Time: 1.244s Load: 0.007s, Pack+Encode: 0.417s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0220 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample131-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample131-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample137-layer4-item1.zst (20/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample137-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 98, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,456B, BPFP=3.8730 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,460B, BPFP=2.9578 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,248B, BPFP=10.3145 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,956B, BPFP=3.9646 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,512B, BPFP=11.5996 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,396B, BPFP=3.6863 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,136B, BPFP=11.5078 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,248B, BPFP=3.4941 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,928B, BPFP=10.4805 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 122,952B, BPFP=2.4504 +⌛️ [2/4] FRONTEND: Frontend time: 0.421s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.826s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000018 0.00007601 + layer.1.conv_state 0.00050366 0.40323901 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00012362 0.07115332 + layer.3.ssm_state 0.00000001 0.00000468 + layer.3.conv_state 0.00007017 0.06279068 + layer.4.ssm_state 0.00000002 0.00000621 + layer.4.conv_state 0.00024945 0.13885270 + layer.4.output 0.00000327 0.00066638 + ------------------------------------------------------------------------------------- + TOTAL 0.00003044 0.02189656 + (elements=1,220,608) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1220608 +Total Bytes 639916 +BPFP 4.1941 bits/point +EBPFP 7.5823 equivalent bits/point +MSE 0.021897 +---------------------- -------------------------------------------------------- +Time: 1.255s Load: 0.008s, Pack+Encode: 0.421s, Decode+Unpack: 0.826s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0219 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample137-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample137-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample139-layer4-item1.zst (21/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample139-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 98, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,496B, BPFP=3.8755 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,452B, BPFP=2.9573 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,216B, BPFP=10.3066 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,200B, BPFP=3.9795 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,512B, BPFP=11.5996 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,864B, BPFP=3.7148 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,168B, BPFP=11.5156 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,248B, BPFP=3.4941 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,872B, BPFP=10.4668 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,036B, BPFP=2.4521 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007405 + layer.1.conv_state 0.00050590 0.40136033 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013813 0.07096085 + layer.3.ssm_state 0.00000001 0.00000461 + layer.3.conv_state 0.00006868 0.06258652 + layer.4.ssm_state 0.00000003 0.00000606 + layer.4.conv_state 0.00022950 0.13773432 + layer.4.output 0.00000311 0.00067894 + ------------------------------------------------------------------------------------- + TOTAL 0.00003028 0.02180935 + (elements=1,220,608) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1220608 +Total Bytes 640688 +BPFP 4.1991 bits/point +EBPFP 7.5919 equivalent bits/point +MSE 0.021809 +---------------------- -------------------------------------------------------- +Time: 1.245s Load: 0.008s, Pack+Encode: 0.417s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample139-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample139-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample14-layer4-item1.zst (22/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample14-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 106, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,512B, BPFP=3.8765 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,564B, BPFP=2.9641 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,172B, BPFP=10.2959 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,108B, BPFP=3.9739 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,388B, BPFP=11.5693 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,376B, BPFP=3.6851 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,044B, BPFP=11.4854 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,472B, BPFP=3.5078 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,832B, BPFP=10.4570 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 131,496B, BPFP=2.4229 +⌛️ [2/4] FRONTEND: Frontend time: 0.423s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.824s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007243 + layer.1.conv_state 0.00051424 0.40107560 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013151 0.07073951 + layer.3.ssm_state 0.00000001 0.00000454 + layer.3.conv_state 0.00006895 0.06264078 + layer.4.ssm_state 0.00000001 0.00000599 + layer.4.conv_state 0.00023185 0.13460228 + layer.4.output 0.00000293 0.00057752 + ------------------------------------------------------------------------------------- + TOTAL 0.00002960 0.02112791 + (elements=1,253,376) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1253376 +Total Bytes 648588 +BPFP 4.1398 bits/point +EBPFP 7.4403 equivalent bits/point +MSE 0.021128 +---------------------- -------------------------------------------------------- +Time: 1.256s Load: 0.009s, Pack+Encode: 0.423s, Decode+Unpack: 0.824s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0211 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample14-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample14-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample149-layer4-item1.zst (23/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample149-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,316B, BPFP=3.8645 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,660B, BPFP=2.9700 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,200B, BPFP=10.3027 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,016B, BPFP=3.9683 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,496B, BPFP=11.5957 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,884B, BPFP=3.7161 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,144B, BPFP=11.5098 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,452B, BPFP=3.5066 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,940B, BPFP=10.4834 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,300B, BPFP=2.4139 +⌛️ [2/4] FRONTEND: Frontend time: 0.440s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.823s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007296 + layer.1.conv_state 0.00051274 0.40369099 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00014729 0.07126256 + layer.3.ssm_state 0.00000001 0.00000481 + layer.3.conv_state 0.00006977 0.06308670 + layer.4.ssm_state 0.00000001 0.00000591 + layer.4.conv_state 0.00021673 0.13482675 + layer.4.output 0.00000307 0.00065834 + ------------------------------------------------------------------------------------- + TOTAL 0.00002991 0.02145949 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 645032 +BPFP 4.1578 bits/point +EBPFP 7.4951 equivalent bits/point +MSE 0.021459 +---------------------- -------------------------------------------------------- +Time: 1.271s Load: 0.008s, Pack+Encode: 0.440s, Decode+Unpack: 0.823s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample149-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample149-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample15-layer4-item1.zst (24/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample15-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,364B, BPFP=3.8674 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,524B, BPFP=2.9617 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,288B, BPFP=10.3242 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,200B, BPFP=3.9795 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,512B, BPFP=11.5996 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,828B, BPFP=3.7126 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,240B, BPFP=11.5332 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,912B, BPFP=3.4736 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,956B, BPFP=10.4873 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,484B, BPFP=2.4411 +⌛️ [2/4] FRONTEND: Frontend time: 0.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.823s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007219 + layer.1.conv_state 0.00050701 0.40396473 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00014784 0.07101651 + layer.3.ssm_state 0.00000001 0.00000471 + layer.3.conv_state 0.00006861 0.06282824 + layer.4.ssm_state 0.00000001 0.00000617 + layer.4.conv_state 0.00022948 0.14338124 + layer.4.output 0.00000294 0.00065735 + ------------------------------------------------------------------------------------- + TOTAL 0.00003012 0.02174846 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 644932 +BPFP 4.1710 bits/point +EBPFP 7.5175 equivalent bits/point +MSE 0.021748 +---------------------- -------------------------------------------------------- +Time: 1.267s Load: 0.009s, Pack+Encode: 0.435s, Decode+Unpack: 0.823s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0217 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample15-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample15-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample153-layer4-item1.zst (25/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample153-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 104, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,352B, BPFP=3.8667 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,420B, BPFP=2.9553 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,200B, BPFP=10.3027 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,980B, BPFP=3.9661 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,600B, BPFP=11.6211 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,656B, BPFP=3.7021 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,224B, BPFP=11.5293 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,264B, BPFP=3.4951 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,088B, BPFP=10.5195 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,180B, BPFP=2.3884 +⌛️ [2/4] FRONTEND: Frontend time: 0.423s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.824s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007246 + layer.1.conv_state 0.00050709 0.40380844 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00011582 0.07119542 + layer.3.ssm_state 0.00000001 0.00000470 + layer.3.conv_state 0.00007623 0.06286700 + layer.4.ssm_state 0.00000002 0.00000598 + layer.4.conv_state 0.00025645 0.13397093 + layer.4.output 0.00000303 0.00049624 + ------------------------------------------------------------------------------------- + TOTAL 0.00003006 0.02130857 + (elements=1,245,184) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1245184 +Total Bytes 644588 +BPFP 4.1413 bits/point +EBPFP 7.4655 equivalent bits/point +MSE 0.021309 +---------------------- -------------------------------------------------------- +Time: 1.256s Load: 0.009s, Pack+Encode: 0.423s, Decode+Unpack: 0.824s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0213 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample153-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample153-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample16-layer4-item1.zst (26/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample16-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,516B, BPFP=3.8767 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,612B, BPFP=2.9670 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,132B, BPFP=3.9753 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,600B, BPFP=11.6211 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,500B, BPFP=3.6926 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,252B, BPFP=11.5361 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,948B, BPFP=3.4758 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,812B, BPFP=10.4521 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,964B, BPFP=2.4602 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.816s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000246 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007359 + layer.1.conv_state 0.00050979 0.40278435 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00012394 0.07053054 + layer.3.ssm_state 0.00000001 0.00000471 + layer.3.conv_state 0.00006896 0.06321947 + layer.4.ssm_state 0.00000002 0.00000650 + layer.4.conv_state 0.00026177 0.14412510 + layer.4.output 0.00000303 0.00053862 + ------------------------------------------------------------------------------------- + TOTAL 0.00003064 0.02183551 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 643168 +BPFP 4.1873 bits/point +EBPFP 7.5545 equivalent bits/point +MSE 0.021836 +---------------------- -------------------------------------------------------- +Time: 1.241s Load: 0.007s, Pack+Encode: 0.418s, Decode+Unpack: 0.816s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample16-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample16-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample168-layer4-item1.zst (27/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample168-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,456B, BPFP=3.8730 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,664B, BPFP=2.9702 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,196B, BPFP=10.3018 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,064B, BPFP=3.9712 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,448B, BPFP=11.5840 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,992B, BPFP=3.7227 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,312B, BPFP=11.5508 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,500B, BPFP=3.5095 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,836B, BPFP=10.4580 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 126,792B, BPFP=2.4764 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.827s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007438 + layer.1.conv_state 0.00051051 0.40244365 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00012244 0.07108021 + layer.3.ssm_state 0.00000001 0.00000474 + layer.3.conv_state 0.00006890 0.06283476 + layer.4.ssm_state 0.00000001 0.00000586 + layer.4.conv_state 0.00024994 0.13152541 + layer.4.output 0.00000315 0.00053304 + ------------------------------------------------------------------------------------- + TOTAL 0.00003034 0.02149300 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 644884 +BPFP 4.1985 bits/point +EBPFP 7.5715 equivalent bits/point +MSE 0.021493 +---------------------- -------------------------------------------------------- +Time: 1.255s Load: 0.008s, Pack+Encode: 0.419s, Decode+Unpack: 0.827s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample168-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample168-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample179-layer4-item1.zst (28/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample179-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 105, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,552B, BPFP=3.8789 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,640B, BPFP=2.9688 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,276B, BPFP=10.3213 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,184B, BPFP=3.9785 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,568B, BPFP=11.6133 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,748B, BPFP=3.7078 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,220B, BPFP=11.5283 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,312B, BPFP=3.4980 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,884B, BPFP=10.4697 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 129,148B, BPFP=2.4023 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.819s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007499 + layer.1.conv_state 0.00051452 0.40441218 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00013443 0.07127167 + layer.3.ssm_state 0.00000001 0.00000461 + layer.3.conv_state 0.00007113 0.06295731 + layer.4.ssm_state 0.00000002 0.00000596 + layer.4.conv_state 0.00021975 0.13800874 + layer.4.output 0.00000261 0.00049413 + ------------------------------------------------------------------------------------- + TOTAL 0.00002940 0.02136598 + (elements=1,249,280) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1249280 +Total Bytes 647156 +BPFP 4.1442 bits/point +EBPFP 7.4613 equivalent bits/point +MSE 0.021366 +---------------------- -------------------------------------------------------- +Time: 1.245s Load: 0.008s, Pack+Encode: 0.418s, Decode+Unpack: 0.819s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample179-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample179-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample180-layer4-item1.zst (29/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample180-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 105, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,560B, BPFP=3.8794 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,684B, BPFP=2.9714 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,232B, BPFP=10.3105 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,232B, BPFP=3.9814 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,568B, BPFP=11.6133 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,588B, BPFP=3.6980 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,168B, BPFP=11.5156 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,156B, BPFP=3.4885 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,864B, BPFP=10.4648 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 129,048B, BPFP=2.4004 +⌛️ [2/4] FRONTEND: Frontend time: 0.424s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007491 + layer.1.conv_state 0.00051408 0.40414932 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00013299 0.07112090 + layer.3.ssm_state 0.00000001 0.00000466 + layer.3.conv_state 0.00006990 0.06344035 + layer.4.ssm_state 0.00000003 0.00000603 + layer.4.conv_state 0.00024394 0.14059284 + layer.4.output 0.00000276 0.00049161 + ------------------------------------------------------------------------------------- + TOTAL 0.00003002 0.02143471 + (elements=1,249,280) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1249280 +Total Bytes 646724 +BPFP 4.1414 bits/point +EBPFP 7.4565 equivalent bits/point +MSE 0.021435 +---------------------- -------------------------------------------------------- +Time: 1.256s Load: 0.009s, Pack+Encode: 0.424s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample180-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample180-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample183-layer4-item1.zst (30/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample183-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 98, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,356B, BPFP=3.8669 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,620B, BPFP=2.9675 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,224B, BPFP=10.3086 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,076B, BPFP=3.9719 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,644B, BPFP=11.6318 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,656B, BPFP=3.7021 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,084B, BPFP=11.4951 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,724B, BPFP=3.5232 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,976B, BPFP=10.4922 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,208B, BPFP=2.4954 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 98, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000029 0.00007336 + layer.1.conv_state 0.00052074 0.40295330 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00013395 0.07168569 + layer.3.ssm_state 0.00000001 0.00000471 + layer.3.conv_state 0.00007102 0.06325638 + layer.4.ssm_state 0.00000002 0.00000568 + layer.4.conv_state 0.00024289 0.12934633 + layer.4.output 0.00000300 0.00065847 + ------------------------------------------------------------------------------------- + TOTAL 0.00003095 0.02165754 + (elements=1,220,608) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1220608 +Total Bytes 643192 +BPFP 4.2156 bits/point +EBPFP 7.6105 equivalent bits/point +MSE 0.021658 +---------------------- -------------------------------------------------------- +Time: 1.246s Load: 0.007s, Pack+Encode: 0.417s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0217 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample183-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample183-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample185-layer4-item1.zst (31/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample185-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 99, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,532B, BPFP=3.8777 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,604B, BPFP=2.9666 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,252B, BPFP=10.3154 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,080B, BPFP=3.9722 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,616B, BPFP=11.6250 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,736B, BPFP=3.7070 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,176B, BPFP=11.5176 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,696B, BPFP=3.5215 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,904B, BPFP=10.4746 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 124,928B, BPFP=2.4646 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.825s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007387 + layer.1.conv_state 0.00050627 0.40240079 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00013355 0.07123026 + layer.3.ssm_state 0.00000001 0.00000471 + layer.3.conv_state 0.00006878 0.06321843 + layer.4.ssm_state 0.00000005 0.00000577 + layer.4.conv_state 0.00026388 0.13090289 + layer.4.output 0.00000306 0.00065756 + ------------------------------------------------------------------------------------- + TOTAL 0.00003096 0.02160073 + (elements=1,224,704) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1224704 +Total Bytes 643148 +BPFP 4.2012 bits/point +EBPFP 7.5863 equivalent bits/point +MSE 0.021601 +---------------------- -------------------------------------------------------- +Time: 1.251s Load: 0.007s, Pack+Encode: 0.419s, Decode+Unpack: 0.825s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0216 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample185-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample185-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample187-layer4-item1.zst (32/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample187-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,460B, BPFP=3.8733 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,624B, BPFP=2.9678 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,216B, BPFP=10.3066 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,084B, BPFP=3.9724 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,624B, BPFP=11.6270 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,764B, BPFP=3.7087 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,180B, BPFP=11.5186 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,372B, BPFP=3.5017 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,872B, BPFP=10.4668 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,424B, BPFP=2.4106 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.832s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007304 + layer.1.conv_state 0.00050939 0.40249798 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00012181 0.07069568 + layer.3.ssm_state 0.00000001 0.00000475 + layer.3.conv_state 0.00007190 0.06287514 + layer.4.ssm_state 0.00000003 0.00000619 + layer.4.conv_state 0.00024002 0.13921848 + layer.4.output 0.00000295 0.00053346 + ------------------------------------------------------------------------------------- + TOTAL 0.00003005 0.02169045 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 641244 +BPFP 4.1748 bits/point +EBPFP 7.5460 equivalent bits/point +MSE 0.021690 +---------------------- -------------------------------------------------------- +Time: 1.256s Load: 0.007s, Pack+Encode: 0.417s, Decode+Unpack: 0.832s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0217 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample187-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample187-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample19-layer4-item1.zst (33/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample19-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,456B, BPFP=3.8730 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,540B, BPFP=2.9626 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,172B, BPFP=10.2959 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,144B, BPFP=3.9761 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,560B, BPFP=11.6113 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,436B, BPFP=3.6887 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,180B, BPFP=11.5186 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,056B, BPFP=3.4824 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,816B, BPFP=10.4531 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,940B, BPFP=2.4598 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.826s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007353 + layer.1.conv_state 0.00050793 0.40258420 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00012495 0.07066094 + layer.3.ssm_state 0.00000001 0.00000468 + layer.3.conv_state 0.00006861 0.06342293 + layer.4.ssm_state 0.00000002 0.00000634 + layer.4.conv_state 0.00029770 0.14281411 + layer.4.output 0.00000302 0.00056092 + ------------------------------------------------------------------------------------- + TOTAL 0.00003157 0.02181152 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 642924 +BPFP 4.1857 bits/point +EBPFP 7.5515 equivalent bits/point +MSE 0.021812 +---------------------- -------------------------------------------------------- +Time: 1.252s Load: 0.008s, Pack+Encode: 0.418s, Decode+Unpack: 0.826s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample19-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample19-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample196-layer4-item1.zst (34/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample196-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,560B, BPFP=3.8794 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,544B, BPFP=2.9629 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,220B, BPFP=10.3076 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,040B, BPFP=3.9697 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,528B, BPFP=11.6035 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,528B, BPFP=3.6943 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,256B, BPFP=11.5371 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,544B, BPFP=3.5122 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,004B, BPFP=10.4990 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,120B, BPFP=2.4438 +⌛️ [2/4] FRONTEND: Frontend time: 0.425s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.826s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000029 0.00007398 + layer.1.conv_state 0.00050905 0.40216070 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00011183 0.07077122 + layer.3.ssm_state 0.00000001 0.00000470 + layer.3.conv_state 0.00006817 0.06264544 + layer.4.ssm_state 0.00000005 0.00000583 + layer.4.conv_state 0.00022612 0.13137299 + layer.4.output 0.00000307 0.00055539 + ------------------------------------------------------------------------------------- + TOTAL 0.00002936 0.02147550 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 642968 +BPFP 4.1860 bits/point +EBPFP 7.5574 equivalent bits/point +MSE 0.021475 +---------------------- -------------------------------------------------------- +Time: 1.259s Load: 0.008s, Pack+Encode: 0.425s, Decode+Unpack: 0.826s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample196-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample196-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample2-layer4-item1.zst (35/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample2-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 105, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,384B, BPFP=3.8687 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,572B, BPFP=2.9646 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,052B, BPFP=3.9705 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,468B, BPFP=11.5889 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,320B, BPFP=3.6816 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,136B, BPFP=11.5078 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,416B, BPFP=3.5044 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,924B, BPFP=10.4795 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 131,396B, BPFP=2.4441 +⌛️ [2/4] FRONTEND: Frontend time: 0.416s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.824s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007304 + layer.1.conv_state 0.00049482 0.40231112 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013819 0.07120093 + layer.3.ssm_state 0.00000001 0.00000454 + layer.3.conv_state 0.00007115 0.06297763 + layer.4.ssm_state 0.00000002 0.00000609 + layer.4.conv_state 0.00022532 0.13469619 + layer.4.output 0.00000291 0.00049519 + ------------------------------------------------------------------------------------- + TOTAL 0.00002923 0.02122283 + (elements=1,249,280) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1249280 +Total Bytes 648500 +BPFP 4.1528 bits/point +EBPFP 7.4642 equivalent bits/point +MSE 0.021223 +---------------------- -------------------------------------------------------- +Time: 1.249s Load: 0.009s, Pack+Encode: 0.416s, Decode+Unpack: 0.824s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0212 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample2-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample2-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample21-layer4-item1.zst (36/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample21-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 104, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,492B, BPFP=3.8752 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,644B, BPFP=2.9690 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,228B, BPFP=10.3096 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,120B, BPFP=3.9746 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,500B, BPFP=11.5967 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,424B, BPFP=3.6880 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,020B, BPFP=11.4795 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,344B, BPFP=3.5000 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,872B, BPFP=10.4668 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,788B, BPFP=2.3999 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.825s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007125 + layer.1.conv_state 0.00050941 0.40266168 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00014139 0.07083334 + layer.3.ssm_state 0.00000001 0.00000462 + layer.3.conv_state 0.00007401 0.06260203 + layer.4.ssm_state 0.00000002 0.00000591 + layer.4.conv_state 0.00025094 0.13581198 + layer.4.output 0.00000282 0.00050608 + ------------------------------------------------------------------------------------- + TOTAL 0.00003050 0.02131357 + (elements=1,245,184) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1245184 +Total Bytes 645056 +BPFP 4.1443 bits/point +EBPFP 7.4676 equivalent bits/point +MSE 0.021314 +---------------------- -------------------------------------------------------- +Time: 1.254s Load: 0.009s, Pack+Encode: 0.420s, Decode+Unpack: 0.825s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0213 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample21-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample21-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample22-layer4-item1.zst (37/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample22-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 104, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,620B, BPFP=3.8831 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,640B, BPFP=2.9688 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,160B, BPFP=10.2930 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,148B, BPFP=3.9763 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,456B, BPFP=11.5859 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,588B, BPFP=3.6980 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,124B, BPFP=11.5049 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,196B, BPFP=3.4910 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,744B, BPFP=10.4355 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 129,996B, BPFP=2.4413 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.821s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007254 + layer.1.conv_state 0.00050400 0.40273815 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013267 0.07039788 + layer.3.ssm_state 0.00000001 0.00000455 + layer.3.conv_state 0.00007191 0.06246903 + layer.4.ssm_state 0.00000002 0.00000599 + layer.4.conv_state 0.00022821 0.13698681 + layer.4.output 0.00000291 0.00052175 + ------------------------------------------------------------------------------------- + TOTAL 0.00002951 0.02133704 + (elements=1,245,184) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1245184 +Total Bytes 647296 +BPFP 4.1587 bits/point +EBPFP 7.4822 equivalent bits/point +MSE 0.021337 +---------------------- -------------------------------------------------------- +Time: 1.247s Load: 0.008s, Pack+Encode: 0.418s, Decode+Unpack: 0.821s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0213 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample22-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample22-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample220-layer4-item1.zst (38/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample220-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 101, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,408B, BPFP=3.8701 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,656B, BPFP=2.9697 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,228B, BPFP=10.3096 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,092B, BPFP=3.9729 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,572B, BPFP=11.6143 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,516B, BPFP=3.6936 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,008B, BPFP=11.4766 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,128B, BPFP=3.4868 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,984B, BPFP=10.4941 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,680B, BPFP=2.3917 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.835s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007249 + layer.1.conv_state 0.00051497 0.40325284 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00014786 0.07024411 + layer.3.ssm_state 0.00000001 0.00000472 + layer.3.conv_state 0.00006809 0.06322186 + layer.4.ssm_state 0.00000002 0.00000611 + layer.4.conv_state 0.00024536 0.13901508 + layer.4.output 0.00000298 0.00052178 + ------------------------------------------------------------------------------------- + TOTAL 0.00003085 0.02162804 + (elements=1,232,896) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1232896 +Total Bytes 640896 +BPFP 4.1586 bits/point +EBPFP 7.5147 equivalent bits/point +MSE 0.021628 +---------------------- -------------------------------------------------------- +Time: 1.261s Load: 0.009s, Pack+Encode: 0.417s, Decode+Unpack: 0.835s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0216 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample220-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample220-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample23-layer4-item1.zst (39/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample23-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 109, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 109, 4096]) -> torch.Size([1, 1, 109, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,568B, BPFP=3.8799 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,556B, BPFP=2.9636 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,248B, BPFP=10.3145 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,052B, BPFP=3.9705 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,556B, BPFP=11.6104 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,868B, BPFP=3.7151 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,164B, BPFP=11.5146 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,996B, BPFP=3.4788 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,892B, BPFP=10.4717 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 134,624B, BPFP=2.4123 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 109, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.824s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 109, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007434 + layer.1.conv_state 0.00052078 0.40426826 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00012449 0.07160592 + layer.3.ssm_state 0.00000001 0.00000465 + layer.3.conv_state 0.00006985 0.06345069 + layer.4.ssm_state 0.00000002 0.00000597 + layer.4.conv_state 0.00025944 0.13699557 + layer.4.output 0.00000291 0.00049731 + ------------------------------------------------------------------------------------- + TOTAL 0.00003006 0.02108833 + (elements=1,265,664) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1265664 +Total Bytes 652148 +BPFP 4.1221 bits/point +EBPFP 7.3933 equivalent bits/point +MSE 0.021088 +---------------------- -------------------------------------------------------- +Time: 1.252s Load: 0.008s, Pack+Encode: 0.420s, Decode+Unpack: 0.824s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0211 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample23-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample23-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample236-layer4-item1.zst (40/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample236-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 96, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 96, 4096]) -> torch.Size([1, 1, 96, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,504B, BPFP=3.8760 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,624B, BPFP=2.9678 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,224B, BPFP=10.3086 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,132B, BPFP=3.9753 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,620B, BPFP=11.6260 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,544B, BPFP=3.6953 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,224B, BPFP=11.5293 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,908B, BPFP=3.4734 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,972B, BPFP=10.4912 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 120,844B, BPFP=2.4586 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 96, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.827s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 96, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007325 + layer.1.conv_state 0.00050897 0.40189615 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00013010 0.07154085 + layer.3.ssm_state 0.00000001 0.00000475 + layer.3.conv_state 0.00007001 0.06337008 + layer.4.ssm_state 0.00000002 0.00000602 + layer.4.conv_state 0.00022365 0.13519996 + layer.4.output 0.00000328 0.00059752 + ------------------------------------------------------------------------------------- + TOTAL 0.00003024 0.02190848 + (elements=1,212,416) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1212416 +Total Bytes 638220 +BPFP 4.2112 bits/point +EBPFP 7.6251 equivalent bits/point +MSE 0.021908 +---------------------- -------------------------------------------------------- +Time: 1.253s Load: 0.008s, Pack+Encode: 0.418s, Decode+Unpack: 0.827s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0219 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample236-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample236-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample24-layer4-item1.zst (41/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample24-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 101, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,496B, BPFP=3.8755 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,652B, BPFP=2.9695 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,228B, BPFP=10.3096 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,108B, BPFP=3.9739 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,472B, BPFP=11.5898 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,544B, BPFP=3.6953 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 46,872B, BPFP=11.4434 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,524B, BPFP=3.5110 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,744B, BPFP=10.4355 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,520B, BPFP=2.4273 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.819s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007383 + layer.1.conv_state 0.00050621 0.40240851 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00012991 0.07066020 + layer.3.ssm_state 0.00000001 0.00000449 + layer.3.conv_state 0.00007072 0.06244703 + layer.4.ssm_state 0.00000001 0.00000586 + layer.4.conv_state 0.00026898 0.13305579 + layer.4.output 0.00000300 0.00053485 + ------------------------------------------------------------------------------------- + TOTAL 0.00003084 0.02144216 + (elements=1,232,896) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1232896 +Total Bytes 642784 +BPFP 4.1709 bits/point +EBPFP 7.5273 equivalent bits/point +MSE 0.021442 +---------------------- -------------------------------------------------------- +Time: 1.246s Load: 0.007s, Pack+Encode: 0.419s, Decode+Unpack: 0.819s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample24-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample24-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample28-layer4-item1.zst (42/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample28-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 106, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,464B, BPFP=3.8735 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,648B, BPFP=2.9692 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,196B, BPFP=10.3018 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,108B, BPFP=3.9739 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,464B, BPFP=11.5879 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,764B, BPFP=3.7087 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,164B, BPFP=11.5146 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,432B, BPFP=3.5054 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,800B, BPFP=10.4492 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 131,968B, BPFP=2.4316 +⌛️ [2/4] FRONTEND: Frontend time: 0.424s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.817s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007320 + layer.1.conv_state 0.00051084 0.40306163 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00012173 0.07058375 + layer.3.ssm_state 0.00000001 0.00000471 + layer.3.conv_state 0.00007090 0.06265950 + layer.4.ssm_state 0.00000002 0.00000593 + layer.4.conv_state 0.00021893 0.13474454 + layer.4.output 0.00000300 0.00058577 + ------------------------------------------------------------------------------------- + TOTAL 0.00002901 0.02118292 + (elements=1,253,376) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1253376 +Total Bytes 649632 +BPFP 4.1464 bits/point +EBPFP 7.4506 equivalent bits/point +MSE 0.021183 +---------------------- -------------------------------------------------------- +Time: 1.249s Load: 0.009s, Pack+Encode: 0.424s, Decode+Unpack: 0.817s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0212 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample28-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample28-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample29-layer4-item1.zst (43/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample29-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,520B, BPFP=3.8770 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,664B, BPFP=2.9702 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,240B, BPFP=10.3125 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,044B, BPFP=3.9700 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,556B, BPFP=11.6104 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,880B, BPFP=3.7158 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,244B, BPFP=11.5342 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,248B, BPFP=3.4941 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,976B, BPFP=10.4922 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,220B, BPFP=2.3977 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 1.261s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007343 + layer.1.conv_state 0.00051310 0.40312833 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00013079 0.07144505 + layer.3.ssm_state 0.00000001 0.00000470 + layer.3.conv_state 0.00006987 0.06340019 + layer.4.ssm_state 0.00000001 0.00000591 + layer.4.conv_state 0.00025978 0.13379383 + layer.4.output 0.00000300 0.00064159 + ------------------------------------------------------------------------------------- + TOTAL 0.00003069 0.02149362 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 643216 +BPFP 4.1599 bits/point +EBPFP 7.5099 equivalent bits/point +MSE 0.021494 +---------------------- -------------------------------------------------------- +Time: 1.690s Load: 0.009s, Pack+Encode: 0.419s, Decode+Unpack: 1.261s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample29-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample29-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample30-layer4-item1.zst (44/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample30-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 108, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 108, 4096]) -> torch.Size([1, 1, 108, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,520B, BPFP=3.8770 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,588B, BPFP=2.9656 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,232B, BPFP=10.3105 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,064B, BPFP=3.9712 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,536B, BPFP=11.6055 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,692B, BPFP=3.7043 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,200B, BPFP=11.5234 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,244B, BPFP=3.4939 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,016B, BPFP=10.5020 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 132,448B, BPFP=2.3953 +⌛️ [2/4] FRONTEND: Frontend time: 0.521s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 108, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.953s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 108, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007218 + layer.1.conv_state 0.00051348 0.40309232 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00012963 0.07148284 + layer.3.ssm_state 0.00000001 0.00000478 + layer.3.conv_state 0.00007448 0.06326593 + layer.4.ssm_state 0.00000002 0.00000599 + layer.4.conv_state 0.00025769 0.13966569 + layer.4.output 0.00000283 0.00050547 + ------------------------------------------------------------------------------------- + TOTAL 0.00003014 0.02118865 + (elements=1,261,568) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1261568 +Total Bytes 650164 +BPFP 4.1229 bits/point +EBPFP 7.4059 equivalent bits/point +MSE 0.021189 +---------------------- -------------------------------------------------------- +Time: 1.484s Load: 0.010s, Pack+Encode: 0.521s, Decode+Unpack: 0.953s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0212 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample30-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample30-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample31-layer4-item1.zst (45/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample31-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 107, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 107, 4096]) -> torch.Size([1, 1, 107, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,532B, BPFP=3.8777 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,584B, BPFP=2.9653 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,164B, BPFP=10.2939 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,164B, BPFP=3.9773 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,468B, BPFP=11.5889 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,528B, BPFP=3.6943 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,012B, BPFP=11.4775 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,572B, BPFP=3.5139 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,716B, BPFP=10.4287 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 131,840B, BPFP=2.4065 +⌛️ [2/4] FRONTEND: Frontend time: 0.428s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 107, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 107, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007272 + layer.1.conv_state 0.00050435 0.40052170 + layer.2.ssm_state 0.00000001 0.00000367 + layer.2.conv_state 0.00013424 0.07030230 + layer.3.ssm_state 0.00000001 0.00000454 + layer.3.conv_state 0.00007347 0.06225724 + layer.4.ssm_state 0.00000005 0.00000608 + layer.4.conv_state 0.00029085 0.13100491 + layer.4.output 0.00000286 0.00060478 + ------------------------------------------------------------------------------------- + TOTAL 0.00003097 0.02094095 + (elements=1,257,472) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1257472 +Total Bytes 649204 +BPFP 4.1302 bits/point +EBPFP 7.4217 equivalent bits/point +MSE 0.020941 +---------------------- -------------------------------------------------------- +Time: 1.254s Load: 0.008s, Pack+Encode: 0.428s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0209 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample31-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample31-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample32-layer4-item1.zst (46/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample32-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 105, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,528B, BPFP=3.8774 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,540B, BPFP=2.9626 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,248B, BPFP=10.3145 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,108B, BPFP=3.9739 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,572B, BPFP=11.6143 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,796B, BPFP=3.7107 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,164B, BPFP=11.5146 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,332B, BPFP=3.4993 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,928B, BPFP=10.4805 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,004B, BPFP=2.3810 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.827s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007388 + layer.1.conv_state 0.00051859 0.40305910 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013121 0.07098603 + layer.3.ssm_state 0.00000001 0.00000481 + layer.3.conv_state 0.00006790 0.06288438 + layer.4.ssm_state 0.00000004 0.00000589 + layer.4.conv_state 0.00024976 0.13328269 + layer.4.output 0.00000299 0.00049388 + ------------------------------------------------------------------------------------- + TOTAL 0.00003026 0.02119693 + (elements=1,249,280) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1249280 +Total Bytes 645844 +BPFP 4.1358 bits/point +EBPFP 7.4519 equivalent bits/point +MSE 0.021197 +---------------------- -------------------------------------------------------- +Time: 1.255s Load: 0.010s, Pack+Encode: 0.418s, Decode+Unpack: 0.827s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0212 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample32-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample32-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample33-layer4-item1.zst (47/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample33-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 107, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 107, 4096]) -> torch.Size([1, 1, 107, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,580B, BPFP=3.8806 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,368B, BPFP=2.9521 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,176B, BPFP=10.2969 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,172B, BPFP=3.9778 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,580B, BPFP=11.6162 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,820B, BPFP=3.7122 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,100B, BPFP=11.4990 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,708B, BPFP=3.5222 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,796B, BPFP=10.4482 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 131,300B, BPFP=2.3967 +⌛️ [2/4] FRONTEND: Frontend time: 0.424s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 107, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.825s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 107, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007347 + layer.1.conv_state 0.00050815 0.40099841 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00013662 0.07017083 + layer.3.ssm_state 0.00000001 0.00000469 + layer.3.conv_state 0.00006925 0.06204130 + layer.4.ssm_state 0.00000002 0.00000572 + layer.4.conv_state 0.00024406 0.12885942 + layer.4.output 0.00000278 0.00059823 + ------------------------------------------------------------------------------------- + TOTAL 0.00002977 0.02088618 + (elements=1,257,472) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1257472 +Total Bytes 649224 +BPFP 4.1303 bits/point +EBPFP 7.4254 equivalent bits/point +MSE 0.020886 +---------------------- -------------------------------------------------------- +Time: 1.258s Load: 0.009s, Pack+Encode: 0.424s, Decode+Unpack: 0.825s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0209 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample33-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample33-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample36-layer4-item1.zst (48/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample36-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,464B, BPFP=3.8735 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,596B, BPFP=2.9661 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,180B, BPFP=10.2979 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,208B, BPFP=3.9800 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,516B, BPFP=11.6006 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,920B, BPFP=3.7183 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,208B, BPFP=11.5254 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,104B, BPFP=3.4854 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,888B, BPFP=10.4707 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,000B, BPFP=2.4318 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.826s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007191 + layer.1.conv_state 0.00050924 0.40081823 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00010465 0.07083248 + layer.3.ssm_state 0.00000001 0.00000471 + layer.3.conv_state 0.00006887 0.06261873 + layer.4.ssm_state 0.00000002 0.00000614 + layer.4.conv_state 0.00021814 0.13900240 + layer.4.output 0.00000296 0.00063804 + ------------------------------------------------------------------------------------- + TOTAL 0.00002875 0.02153214 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 644708 +BPFP 4.1695 bits/point +EBPFP 7.5177 equivalent bits/point +MSE 0.021532 +---------------------- -------------------------------------------------------- +Time: 1.252s Load: 0.009s, Pack+Encode: 0.418s, Decode+Unpack: 0.826s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample36-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample36-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample37-layer4-item1.zst (49/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample37-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 106, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,428B, BPFP=3.8713 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,304B, BPFP=2.9482 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,220B, BPFP=10.3076 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,952B, BPFP=3.9644 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,572B, BPFP=11.6143 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,728B, BPFP=3.7065 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,248B, BPFP=11.5352 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,888B, BPFP=3.5332 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,864B, BPFP=10.4648 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 129,384B, BPFP=2.3840 +⌛️ [2/4] FRONTEND: Frontend time: 0.422s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.830s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007313 + layer.1.conv_state 0.00051009 0.40417433 + layer.2.ssm_state 0.00000001 0.00000357 + layer.2.conv_state 0.00012190 0.07103413 + layer.3.ssm_state 0.00000001 0.00000473 + layer.3.conv_state 0.00011307 0.06259733 + layer.4.ssm_state 0.00000002 0.00000554 + layer.4.conv_state 0.00021713 0.12795721 + layer.4.output 0.00000285 0.00059549 + ------------------------------------------------------------------------------------- + TOTAL 0.00003000 0.02104802 + (elements=1,253,376) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1253376 +Total Bytes 647212 +BPFP 4.1310 bits/point +EBPFP 7.4362 equivalent bits/point +MSE 0.021048 +---------------------- -------------------------------------------------------- +Time: 1.260s Load: 0.008s, Pack+Encode: 0.422s, Decode+Unpack: 0.830s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0210 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample37-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample37-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample38-layer4-item1.zst (50/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample38-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 104, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,440B, BPFP=3.8721 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,552B, BPFP=2.9634 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,244B, BPFP=10.3135 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,028B, BPFP=3.9690 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,496B, BPFP=11.5957 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,456B, BPFP=3.6899 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,364B, BPFP=11.5635 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,700B, BPFP=3.5217 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,968B, BPFP=10.4902 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,468B, BPFP=2.3939 +⌛️ [2/4] FRONTEND: Frontend time: 0.428s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007235 + layer.1.conv_state 0.00050753 0.40163815 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00011558 0.07140025 + layer.3.ssm_state 0.00000001 0.00000458 + layer.3.conv_state 0.00006813 0.06327641 + layer.4.ssm_state 0.00000002 0.00000558 + layer.4.conv_state 0.00021580 0.12983026 + layer.4.output 0.00000303 0.00050915 + ------------------------------------------------------------------------------------- + TOTAL 0.00002878 0.02116301 + (elements=1,245,184) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1245184 +Total Bytes 645340 +BPFP 4.1462 bits/point +EBPFP 7.4734 equivalent bits/point +MSE 0.021163 +---------------------- -------------------------------------------------------- +Time: 1.258s Load: 0.009s, Pack+Encode: 0.428s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0212 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample38-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample38-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample39-layer4-item1.zst (51/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample39-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,412B, BPFP=3.8704 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,644B, BPFP=2.9690 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,124B, BPFP=10.2842 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,128B, BPFP=3.9751 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,432B, BPFP=11.5801 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,692B, BPFP=3.7043 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 46,968B, BPFP=11.4668 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,216B, BPFP=3.4922 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,664B, BPFP=10.4160 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,432B, BPFP=2.4164 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.828s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007228 + layer.1.conv_state 0.00050918 0.40179721 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013447 0.07058901 + layer.3.ssm_state 0.00000001 0.00000460 + layer.3.conv_state 0.00006924 0.06270047 + layer.4.ssm_state 0.00000002 0.00000625 + layer.4.conv_state 0.00025518 0.13709471 + layer.4.output 0.00000276 0.00064452 + ------------------------------------------------------------------------------------- + TOTAL 0.00003039 0.02143663 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 644336 +BPFP 4.1534 bits/point +EBPFP 7.4853 equivalent bits/point +MSE 0.021437 +---------------------- -------------------------------------------------------- +Time: 1.257s Load: 0.009s, Pack+Encode: 0.420s, Decode+Unpack: 0.828s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample39-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample39-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample4-layer4-item1.zst (52/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample4-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 110, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 110, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 110, 4096]) -> torch.Size([1, 1, 110, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,524B, BPFP=3.8772 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,628B, BPFP=2.9680 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,092B, BPFP=3.9729 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,568B, BPFP=11.6133 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,724B, BPFP=3.7063 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,148B, BPFP=11.5107 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,192B, BPFP=3.4907 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,952B, BPFP=10.4863 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 136,636B, BPFP=2.4261 +⌛️ [2/4] FRONTEND: Frontend time: 0.421s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 110, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.821s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 110, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007358 + layer.1.conv_state 0.00050842 0.40222201 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013835 0.07112671 + layer.3.ssm_state 0.00000001 0.00000474 + layer.3.conv_state 0.00007010 0.06332299 + layer.4.ssm_state 0.00000002 0.00000604 + layer.4.conv_state 0.00024386 0.13800822 + layer.4.output 0.00000300 0.00057679 + ------------------------------------------------------------------------------------- + TOTAL 0.00002965 0.02100771 + (elements=1,269,760) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1269760 +Total Bytes 654296 +BPFP 4.1223 bits/point +EBPFP 7.3838 equivalent bits/point +MSE 0.021008 +---------------------- -------------------------------------------------------- +Time: 1.251s Load: 0.010s, Pack+Encode: 0.421s, Decode+Unpack: 0.821s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 110, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0210 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample4-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample4-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample40-layer4-item1.zst (53/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample40-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 99, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,344B, BPFP=3.8662 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,584B, BPFP=2.9653 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,256B, BPFP=10.3164 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,996B, BPFP=3.9670 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,652B, BPFP=11.6338 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,936B, BPFP=3.7192 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,244B, BPFP=11.5342 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,128B, BPFP=3.4868 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,976B, BPFP=10.4922 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,088B, BPFP=2.5270 +⌛️ [2/4] FRONTEND: Frontend time: 0.424s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000251 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007367 + layer.1.conv_state 0.00049945 0.40432155 + layer.2.ssm_state 0.00000001 0.00000357 + layer.2.conv_state 0.00012969 0.07147385 + layer.3.ssm_state 0.00000001 0.00000472 + layer.3.conv_state 0.00007002 0.06325222 + layer.4.ssm_state 0.00000003 0.00000614 + layer.4.conv_state 0.00023573 0.14042917 + layer.4.output 0.00000289 0.00067933 + ------------------------------------------------------------------------------------- + TOTAL 0.00002990 0.02192165 + (elements=1,224,704) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1224704 +Total Bytes 645828 +BPFP 4.2187 bits/point +EBPFP 7.6006 equivalent bits/point +MSE 0.021922 +---------------------- -------------------------------------------------------- +Time: 1.253s Load: 0.010s, Pack+Encode: 0.424s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0219 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample40-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample40-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample41-layer4-item1.zst (54/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample41-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,576B, BPFP=3.8804 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,688B, BPFP=2.9717 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,224B, BPFP=10.3086 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,124B, BPFP=3.9749 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,636B, BPFP=11.6299 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,624B, BPFP=3.7002 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,220B, BPFP=11.5283 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,828B, BPFP=3.4685 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,924B, BPFP=10.4795 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 129,096B, BPFP=2.4720 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000246 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007386 + layer.1.conv_state 0.00051118 0.40361410 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00014206 0.07082416 + layer.3.ssm_state 0.00000001 0.00000477 + layer.3.conv_state 0.00007124 0.06319093 + layer.4.ssm_state 0.00000002 0.00000628 + layer.4.conv_state 0.00028525 0.14087994 + layer.4.output 0.00000298 0.00065336 + ------------------------------------------------------------------------------------- + TOTAL 0.00003164 0.02167627 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 646564 +BPFP 4.1815 bits/point +EBPFP 7.5281 equivalent bits/point +MSE 0.021676 +---------------------- -------------------------------------------------------- +Time: 1.245s Load: 0.008s, Pack+Encode: 0.420s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0217 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample41-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample41-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample43-layer4-item1.zst (55/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample43-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,544B, BPFP=3.8784 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,616B, BPFP=2.9673 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,136B, BPFP=10.2871 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,176B, BPFP=3.9780 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,460B, BPFP=11.5869 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,640B, BPFP=3.7012 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,124B, BPFP=11.5049 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,612B, BPFP=3.5164 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,000B, BPFP=10.4980 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,892B, BPFP=2.4198 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.814s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000025 0.00007339 + layer.1.conv_state 0.00050676 0.40202987 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00012238 0.07032359 + layer.3.ssm_state 0.00000001 0.00000479 + layer.3.conv_state 0.00007028 0.06258935 + layer.4.ssm_state 0.00000002 0.00000565 + layer.4.conv_state 0.00023202 0.13178504 + layer.4.output 0.00000314 0.00052834 + ------------------------------------------------------------------------------------- + TOTAL 0.00002981 0.02146047 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 641824 +BPFP 4.1785 bits/point +EBPFP 7.5505 equivalent bits/point +MSE 0.021460 +---------------------- -------------------------------------------------------- +Time: 1.241s Load: 0.008s, Pack+Encode: 0.419s, Decode+Unpack: 0.814s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample43-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample43-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample44-layer4-item1.zst (56/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample44-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,460B, BPFP=3.8733 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,500B, BPFP=2.9602 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,224B, BPFP=10.3086 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,968B, BPFP=3.9653 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,556B, BPFP=11.6104 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,716B, BPFP=3.7058 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,148B, BPFP=11.5107 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,424B, BPFP=3.5049 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,936B, BPFP=10.4824 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 121,328B, BPFP=2.4430 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.819s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007265 + layer.1.conv_state 0.00051299 0.40081397 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00011946 0.07124387 + layer.3.ssm_state 0.00000001 0.00000464 + layer.3.conv_state 0.00007081 0.06256905 + layer.4.ssm_state 0.00000002 0.00000596 + layer.4.conv_state 0.00021939 0.13506067 + layer.4.output 0.00000316 0.00058175 + ------------------------------------------------------------------------------------- + TOTAL 0.00002984 0.02176901 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 638884 +BPFP 4.2014 bits/point +EBPFP 7.6050 equivalent bits/point +MSE 0.021769 +---------------------- -------------------------------------------------------- +Time: 1.245s Load: 0.009s, Pack+Encode: 0.417s, Decode+Unpack: 0.819s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample44-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample44-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample45-layer4-item1.zst (57/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample45-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 107, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 107, 4096]) -> torch.Size([1, 1, 107, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,464B, BPFP=3.8735 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,644B, BPFP=2.9690 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,184B, BPFP=10.2988 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,228B, BPFP=3.9812 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,564B, BPFP=11.6123 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,880B, BPFP=3.7158 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,264B, BPFP=11.5391 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,900B, BPFP=3.4729 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,120B, BPFP=10.5273 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 133,324B, BPFP=2.4336 +⌛️ [2/4] FRONTEND: Frontend time: 0.442s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 107, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.814s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 107, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000029 0.00007457 + layer.1.conv_state 0.00050544 0.40545264 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00013757 0.07129588 + layer.3.ssm_state 0.00000001 0.00000479 + layer.3.conv_state 0.00007295 0.06305452 + layer.4.ssm_state 0.00000001 0.00000656 + layer.4.conv_state 0.00026139 0.14336415 + layer.4.output 0.00000288 0.00061972 + ------------------------------------------------------------------------------------- + TOTAL 0.00003031 0.02144365 + (elements=1,257,472) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1257472 +Total Bytes 651196 +BPFP 4.1429 bits/point +EBPFP 7.4376 equivalent bits/point +MSE 0.021444 +---------------------- -------------------------------------------------------- +Time: 1.265s Load: 0.009s, Pack+Encode: 0.442s, Decode+Unpack: 0.814s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 107, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample45-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample45-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample46-layer4-item1.zst (58/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample46-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 106, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,440B, BPFP=3.8721 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,468B, BPFP=2.9583 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,964B, BPFP=3.9651 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,544B, BPFP=11.6074 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,488B, BPFP=3.6919 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,100B, BPFP=11.4990 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,932B, BPFP=3.5359 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,076B, BPFP=10.5166 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 130,560B, BPFP=2.4057 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007385 + layer.1.conv_state 0.00050804 0.40357023 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00011728 0.07129236 + layer.3.ssm_state 0.00000001 0.00000461 + layer.3.conv_state 0.00007019 0.06267454 + layer.4.ssm_state 0.00000002 0.00000591 + layer.4.conv_state 0.00023224 0.13164736 + layer.4.output 0.00000302 0.00058494 + ------------------------------------------------------------------------------------- + TOTAL 0.00002915 0.02113393 + (elements=1,253,376) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1253376 +Total Bytes 648404 +BPFP 4.1386 bits/point +EBPFP 7.4439 equivalent bits/point +MSE 0.021134 +---------------------- -------------------------------------------------------- +Time: 1.245s Load: 0.009s, Pack+Encode: 0.418s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0211 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample46-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample46-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample47-layer4-item1.zst (59/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample47-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,476B, BPFP=3.8743 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,528B, BPFP=2.9619 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,260B, BPFP=10.3174 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,988B, BPFP=3.9666 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,624B, BPFP=11.6270 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,776B, BPFP=3.7095 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,164B, BPFP=11.5146 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,496B, BPFP=3.5093 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,848B, BPFP=10.4609 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 121,300B, BPFP=2.4424 +⌛️ [2/4] FRONTEND: Frontend time: 0.421s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.828s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007319 + layer.1.conv_state 0.00051689 0.40125346 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00012124 0.07143091 + layer.3.ssm_state 0.00000001 0.00000463 + layer.3.conv_state 0.00007153 0.06275806 + layer.4.ssm_state 0.00000002 0.00000598 + layer.4.conv_state 0.00021058 0.13530523 + layer.4.output 0.00000315 0.00057579 + ------------------------------------------------------------------------------------- + TOTAL 0.00002977 0.02179568 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 639084 +BPFP 4.2027 bits/point +EBPFP 7.6078 equivalent bits/point +MSE 0.021796 +---------------------- -------------------------------------------------------- +Time: 1.258s Load: 0.009s, Pack+Encode: 0.421s, Decode+Unpack: 0.828s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample47-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample47-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample49-layer4-item1.zst (60/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample49-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 99, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,644B, BPFP=3.8845 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,836B, BPFP=2.9807 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,152B, BPFP=10.2910 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,100B, BPFP=3.9734 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,520B, BPFP=11.6016 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,596B, BPFP=3.6985 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,152B, BPFP=11.5117 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,236B, BPFP=3.4934 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,944B, BPFP=10.4844 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,848B, BPFP=2.4828 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007314 + layer.1.conv_state 0.00050257 0.40344846 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00013543 0.07073672 + layer.3.ssm_state 0.00000001 0.00000472 + layer.3.conv_state 0.00007075 0.06300207 + layer.4.ssm_state 0.00000002 0.00000624 + layer.4.conv_state 0.00024175 0.13906458 + layer.4.output 0.00000314 0.00067786 + ------------------------------------------------------------------------------------- + TOTAL 0.00003040 0.02183483 + (elements=1,224,704) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1224704 +Total Bytes 643652 +BPFP 4.2045 bits/point +EBPFP 7.5869 equivalent bits/point +MSE 0.021835 +---------------------- -------------------------------------------------------- +Time: 1.248s Load: 0.009s, Pack+Encode: 0.419s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample49-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample49-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample50-layer4-item1.zst (61/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample50-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,548B, BPFP=3.8787 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,616B, BPFP=2.9673 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,140B, BPFP=3.9758 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,456B, BPFP=11.5859 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,660B, BPFP=3.7024 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,100B, BPFP=11.4990 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,352B, BPFP=3.5005 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,836B, BPFP=10.4580 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,256B, BPFP=2.3984 +⌛️ [2/4] FRONTEND: Frontend time: 0.425s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007245 + layer.1.conv_state 0.00051037 0.40151137 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00013143 0.07073262 + layer.3.ssm_state 0.00000001 0.00000462 + layer.3.conv_state 0.00007554 0.06285238 + layer.4.ssm_state 0.00000002 0.00000604 + layer.4.conv_state 0.00025897 0.13629811 + layer.4.output 0.00000311 0.00064100 + ------------------------------------------------------------------------------------- + TOTAL 0.00003080 0.02148345 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 642796 +BPFP 4.1572 bits/point +EBPFP 7.5042 equivalent bits/point +MSE 0.021483 +---------------------- -------------------------------------------------------- +Time: 1.253s Load: 0.008s, Pack+Encode: 0.425s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample50-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample50-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample51-layer4-item1.zst (62/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample51-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 105, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,616B, BPFP=3.8828 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,732B, BPFP=2.9744 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,240B, BPFP=10.3125 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,152B, BPFP=3.9766 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,524B, BPFP=11.6025 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,656B, BPFP=3.7021 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,188B, BPFP=11.5205 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,788B, BPFP=3.4661 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,916B, BPFP=10.4775 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 130,132B, BPFP=2.4206 +⌛️ [2/4] FRONTEND: Frontend time: 0.416s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.816s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007344 + layer.1.conv_state 0.00051108 0.40282521 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013025 0.07134381 + layer.3.ssm_state 0.00000001 0.00000483 + layer.3.conv_state 0.00006983 0.06315233 + layer.4.ssm_state 0.00000002 0.00000638 + layer.4.conv_state 0.00023909 0.14536554 + layer.4.output 0.00000287 0.00049068 + ------------------------------------------------------------------------------------- + TOTAL 0.00002976 0.02152304 + (elements=1,249,280) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1249280 +Total Bytes 647568 +BPFP 4.1468 bits/point +EBPFP 7.4603 equivalent bits/point +MSE 0.021523 +---------------------- -------------------------------------------------------- +Time: 1.241s Load: 0.009s, Pack+Encode: 0.416s, Decode+Unpack: 0.816s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample51-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample51-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample53-layer4-item1.zst (63/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample53-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,508B, BPFP=3.8762 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,508B, BPFP=2.9607 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,196B, BPFP=10.3018 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,004B, BPFP=3.9675 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,376B, BPFP=11.5664 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,752B, BPFP=3.7080 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 46,988B, BPFP=11.4717 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,784B, BPFP=3.5269 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,888B, BPFP=10.4707 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 117,472B, BPFP=2.3653 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.814s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007210 + layer.1.conv_state 0.00050795 0.40107557 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00011983 0.07066133 + layer.3.ssm_state 0.00000001 0.00000472 + layer.3.conv_state 0.00006952 0.06275530 + layer.4.ssm_state 0.00000002 0.00000574 + layer.4.conv_state 0.00020281 0.13147345 + layer.4.output 0.00000316 0.00056663 + ------------------------------------------------------------------------------------- + TOTAL 0.00002923 0.02166375 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 635100 +BPFP 4.1765 bits/point +EBPFP 7.5805 equivalent bits/point +MSE 0.021664 +---------------------- -------------------------------------------------------- +Time: 1.239s Load: 0.009s, Pack+Encode: 0.417s, Decode+Unpack: 0.814s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0217 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample53-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample53-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample54-layer4-item1.zst (64/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample54-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 105, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,440B, BPFP=3.8721 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,692B, BPFP=2.9719 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,240B, BPFP=10.3125 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,160B, BPFP=3.9771 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,504B, BPFP=11.5977 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,668B, BPFP=3.7029 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 46,972B, BPFP=11.4678 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,560B, BPFP=3.5132 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,864B, BPFP=10.4648 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,980B, BPFP=2.3992 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007151 + layer.1.conv_state 0.00050049 0.40529004 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00013306 0.07080795 + layer.3.ssm_state 0.00000001 0.00000474 + layer.3.conv_state 0.00007056 0.06235763 + layer.4.ssm_state 0.00000001 0.00000599 + layer.4.conv_state 0.00023923 0.13398437 + layer.4.output 0.00000281 0.00047181 + ------------------------------------------------------------------------------------- + TOTAL 0.00002958 0.02124752 + (elements=1,249,280) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1249280 +Total Bytes 646704 +BPFP 4.1413 bits/point +EBPFP 7.4566 equivalent bits/point +MSE 0.021248 +---------------------- -------------------------------------------------------- +Time: 1.243s Load: 0.007s, Pack+Encode: 0.417s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0212 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample54-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample54-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample55-layer4-item1.zst (65/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample55-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,520B, BPFP=3.8770 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,764B, BPFP=2.9763 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,232B, BPFP=10.3105 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,244B, BPFP=3.9822 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,564B, BPFP=11.6123 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,852B, BPFP=3.7141 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,192B, BPFP=11.5215 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,024B, BPFP=3.4805 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,888B, BPFP=10.4707 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 119,600B, BPFP=2.4082 +⌛️ [2/4] FRONTEND: Frontend time: 0.422s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.845s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007354 + layer.1.conv_state 0.00050183 0.40220982 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00013844 0.07065859 + layer.3.ssm_state 0.00000001 0.00000469 + layer.3.conv_state 0.00006878 0.06317858 + layer.4.ssm_state 0.00000002 0.00000621 + layer.4.conv_state 0.00025869 0.13965917 + layer.4.output 0.00000296 0.00057647 + ------------------------------------------------------------------------------------- + TOTAL 0.00003099 0.02192954 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 637504 +BPFP 4.1923 bits/point +EBPFP 7.5982 equivalent bits/point +MSE 0.021930 +---------------------- -------------------------------------------------------- +Time: 1.275s Load: 0.008s, Pack+Encode: 0.422s, Decode+Unpack: 0.845s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0219 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample55-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample55-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample56-layer4-item1.zst (66/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample56-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 104, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,588B, BPFP=3.8811 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,696B, BPFP=2.9722 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,244B, BPFP=10.3135 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,216B, BPFP=3.9805 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,456B, BPFP=11.5859 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,496B, BPFP=3.6924 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,096B, BPFP=11.4980 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,252B, BPFP=3.4944 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,948B, BPFP=10.4854 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,936B, BPFP=2.4214 +⌛️ [2/4] FRONTEND: Frontend time: 0.428s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.816s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000246 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007263 + layer.1.conv_state 0.00050960 0.40105036 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00014477 0.07055338 + layer.3.ssm_state 0.00000001 0.00000476 + layer.3.conv_state 0.00007109 0.06288269 + layer.4.ssm_state 0.00000003 0.00000619 + layer.4.conv_state 0.00024741 0.13882069 + layer.4.output 0.00000297 0.00052394 + ------------------------------------------------------------------------------------- + TOTAL 0.00003048 0.02135666 + (elements=1,245,184) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1245184 +Total Bytes 646552 +BPFP 4.1539 bits/point +EBPFP 7.4795 equivalent bits/point +MSE 0.021357 +---------------------- -------------------------------------------------------- +Time: 1.253s Load: 0.009s, Pack+Encode: 0.428s, Decode+Unpack: 0.816s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample56-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample56-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample57-layer4-item1.zst (67/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample57-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 106, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) -> torch.Size([1, 1, 106, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,380B, BPFP=3.8684 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,544B, BPFP=2.9629 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,200B, BPFP=10.3027 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,028B, BPFP=3.9690 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,596B, BPFP=11.6201 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,564B, BPFP=3.6965 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,276B, BPFP=11.5420 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,644B, BPFP=3.5183 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,908B, BPFP=10.4756 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 131,052B, BPFP=2.4147 +⌛️ [2/4] FRONTEND: Frontend time: 0.426s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 106, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007260 + layer.1.conv_state 0.00051526 0.40416783 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013199 0.07153969 + layer.3.ssm_state 0.00000001 0.00000465 + layer.3.conv_state 0.00007059 0.06286541 + layer.4.ssm_state 0.00000002 0.00000579 + layer.4.conv_state 0.00023257 0.13309096 + layer.4.output 0.00000278 0.00058313 + ------------------------------------------------------------------------------------- + TOTAL 0.00002965 0.02119798 + (elements=1,253,376) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1253376 +Total Bytes 648816 +BPFP 4.1412 bits/point +EBPFP 7.4460 equivalent bits/point +MSE 0.021198 +---------------------- -------------------------------------------------------- +Time: 1.258s Load: 0.010s, Pack+Encode: 0.426s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 106, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0212 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample57-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample57-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample58-layer4-item1.zst (68/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample58-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,464B, BPFP=3.8735 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,548B, BPFP=2.9631 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,252B, BPFP=10.3154 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,120B, BPFP=3.9746 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,484B, BPFP=11.5928 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,896B, BPFP=3.7168 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,132B, BPFP=11.5068 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,324B, BPFP=3.4988 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,844B, BPFP=10.4600 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 126,716B, BPFP=2.4264 +⌛️ [2/4] FRONTEND: Frontend time: 0.432s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007420 + layer.1.conv_state 0.00051277 0.40310186 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00013342 0.07098813 + layer.3.ssm_state 0.00000001 0.00000473 + layer.3.conv_state 0.00006746 0.06281933 + layer.4.ssm_state 0.00000002 0.00000594 + layer.4.conv_state 0.00024674 0.13629051 + layer.4.output 0.00000297 0.00064050 + ------------------------------------------------------------------------------------- + TOTAL 0.00003035 0.02153129 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 644404 +BPFP 4.1676 bits/point +EBPFP 7.5156 equivalent bits/point +MSE 0.021531 +---------------------- -------------------------------------------------------- +Time: 1.261s Load: 0.009s, Pack+Encode: 0.432s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample58-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample58-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample60-layer4-item1.zst (69/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample60-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,392B, BPFP=3.8691 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,608B, BPFP=2.9668 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,076B, BPFP=3.9719 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,448B, BPFP=11.5840 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,508B, BPFP=3.6931 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,064B, BPFP=11.4902 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,764B, BPFP=3.5256 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,832B, BPFP=10.4570 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 118,212B, BPFP=2.3802 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.821s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007257 + layer.1.conv_state 0.00050499 0.40211225 + layer.2.ssm_state 0.00000001 0.00000358 + layer.2.conv_state 0.00013541 0.07078019 + layer.3.ssm_state 0.00000001 0.00000470 + layer.3.conv_state 0.00007246 0.06260229 + layer.4.ssm_state 0.00000002 0.00000583 + layer.4.conv_state 0.00022642 0.13075009 + layer.4.output 0.00000327 0.00057031 + ------------------------------------------------------------------------------------- + TOTAL 0.00003032 0.02167252 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 635736 +BPFP 4.1807 bits/point +EBPFP 7.5840 equivalent bits/point +MSE 0.021673 +---------------------- -------------------------------------------------------- +Time: 1.246s Load: 0.008s, Pack+Encode: 0.417s, Decode+Unpack: 0.821s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0217 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample60-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample60-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample61-layer4-item1.zst (70/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample61-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,324B, BPFP=3.8650 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,564B, BPFP=2.9641 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,260B, BPFP=10.3174 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,952B, BPFP=3.9644 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,588B, BPFP=11.6182 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,600B, BPFP=3.6987 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,040B, BPFP=11.4844 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,592B, BPFP=3.5151 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,144B, BPFP=10.5332 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,916B, BPFP=2.4202 +⌛️ [2/4] FRONTEND: Frontend time: 0.508s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.885s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007276 + layer.1.conv_state 0.00050447 0.40346885 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00013123 0.07144821 + layer.3.ssm_state 0.00000001 0.00000462 + layer.3.conv_state 0.00007682 0.06277895 + layer.4.ssm_state 0.00000002 0.00000614 + layer.4.conv_state 0.00025075 0.13945293 + layer.4.output 0.00000309 0.00054306 + ------------------------------------------------------------------------------------- + TOTAL 0.00003063 0.02174324 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 641604 +BPFP 4.1771 bits/point +EBPFP 7.5475 equivalent bits/point +MSE 0.021743 +---------------------- -------------------------------------------------------- +Time: 1.402s Load: 0.008s, Pack+Encode: 0.508s, Decode+Unpack: 0.885s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0217 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample61-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample61-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample62-layer4-item1.zst (71/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample62-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,324B, BPFP=3.8650 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,524B, BPFP=2.9617 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,236B, BPFP=10.3115 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,940B, BPFP=3.9636 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,468B, BPFP=11.5889 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,424B, BPFP=3.6880 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,060B, BPFP=11.4893 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,432B, BPFP=3.5054 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,008B, BPFP=10.5000 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,724B, BPFP=2.4165 +⌛️ [2/4] FRONTEND: Frontend time: 0.423s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.819s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007257 + layer.1.conv_state 0.00050765 0.40117791 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013519 0.07151555 + layer.3.ssm_state 0.00000001 0.00000461 + layer.3.conv_state 0.00007639 0.06271511 + layer.4.ssm_state 0.00000003 0.00000617 + layer.4.conv_state 0.00024199 0.13935123 + layer.4.output 0.00000307 0.00053354 + ------------------------------------------------------------------------------------- + TOTAL 0.00003057 0.02167634 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 640764 +BPFP 4.1716 bits/point +EBPFP 7.5378 equivalent bits/point +MSE 0.021676 +---------------------- -------------------------------------------------------- +Time: 1.250s Load: 0.009s, Pack+Encode: 0.423s, Decode+Unpack: 0.819s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0217 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample62-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample62-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample63-layer4-item1.zst (72/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample63-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,532B, BPFP=3.8777 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,636B, BPFP=2.9685 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,196B, BPFP=10.3018 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,080B, BPFP=3.9722 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,460B, BPFP=11.5869 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,784B, BPFP=3.7100 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,232B, BPFP=11.5312 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,364B, BPFP=3.5012 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,904B, BPFP=10.4746 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,716B, BPFP=2.4163 +⌛️ [2/4] FRONTEND: Frontend time: 0.416s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.821s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007257 + layer.1.conv_state 0.00051429 0.40270001 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00012088 0.07093723 + layer.3.ssm_state 0.00000001 0.00000482 + layer.3.conv_state 0.00007035 0.06289122 + layer.4.ssm_state 0.00000002 0.00000594 + layer.4.conv_state 0.00025261 0.13631913 + layer.4.output 0.00000308 0.00054772 + ------------------------------------------------------------------------------------- + TOTAL 0.00003049 0.02163007 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 641528 +BPFP 4.1766 bits/point +EBPFP 7.5478 equivalent bits/point +MSE 0.021630 +---------------------- -------------------------------------------------------- +Time: 1.247s Load: 0.010s, Pack+Encode: 0.416s, Decode+Unpack: 0.821s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0216 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample63-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample63-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample65-layer4-item1.zst (73/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample65-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,552B, BPFP=3.8789 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,592B, BPFP=2.9658 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,116B, BPFP=3.9744 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,584B, BPFP=11.6172 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,472B, BPFP=3.6909 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,200B, BPFP=11.5234 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,940B, BPFP=3.4753 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,852B, BPFP=10.4619 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,420B, BPFP=2.4496 +⌛️ [2/4] FRONTEND: Frontend time: 0.424s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.814s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007315 + layer.1.conv_state 0.00051142 0.40327567 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00015094 0.07104949 + layer.3.ssm_state 0.00000001 0.00000476 + layer.3.conv_state 0.00006998 0.06365453 + layer.4.ssm_state 0.00000002 0.00000627 + layer.4.conv_state 0.00021745 0.14143386 + layer.4.output 0.00000300 0.00054171 + ------------------------------------------------------------------------------------- + TOTAL 0.00003024 0.02180325 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 642560 +BPFP 4.1833 bits/point +EBPFP 7.5501 equivalent bits/point +MSE 0.021803 +---------------------- -------------------------------------------------------- +Time: 1.246s Load: 0.008s, Pack+Encode: 0.424s, Decode+Unpack: 0.814s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample65-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample65-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample66-layer4-item1.zst (74/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample66-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 104, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,512B, BPFP=3.8765 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,648B, BPFP=2.9692 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,176B, BPFP=10.2969 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,204B, BPFP=3.9797 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,612B, BPFP=11.6240 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,632B, BPFP=3.7007 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,160B, BPFP=11.5137 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,408B, BPFP=3.5039 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,988B, BPFP=10.4951 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 129,764B, BPFP=2.4370 +⌛️ [2/4] FRONTEND: Frontend time: 0.416s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000026 0.00007365 + layer.1.conv_state 0.00050859 0.40515056 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00015031 0.07104795 + layer.3.ssm_state 0.00000001 0.00000479 + layer.3.conv_state 0.00007130 0.06368046 + layer.4.ssm_state 0.00000001 0.00000613 + layer.4.conv_state 0.00023281 0.13422154 + layer.4.output 0.00000304 0.00052525 + ------------------------------------------------------------------------------------- + TOTAL 0.00003026 0.02137809 + (elements=1,245,184) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1245184 +Total Bytes 647728 +BPFP 4.1615 bits/point +EBPFP 7.4893 equivalent bits/point +MSE 0.021378 +---------------------- -------------------------------------------------------- +Time: 1.246s Load: 0.009s, Pack+Encode: 0.416s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample66-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample66-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample67-layer4-item1.zst (75/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample67-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,592B, BPFP=3.8813 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,780B, BPFP=2.9773 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,268B, BPFP=10.3193 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,200B, BPFP=3.9795 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,600B, BPFP=11.6211 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,456B, BPFP=3.6899 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,092B, BPFP=11.4971 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,024B, BPFP=3.4805 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,872B, BPFP=10.4668 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,124B, BPFP=2.4534 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.821s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000014 0.00007377 + layer.1.conv_state 0.00050679 0.40403053 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013770 0.07080197 + layer.3.ssm_state 0.00000001 0.00000464 + layer.3.conv_state 0.00006920 0.06346639 + layer.4.ssm_state 0.00000002 0.00000617 + layer.4.conv_state 0.00024331 0.14087915 + layer.4.output 0.00000290 0.00065225 + ------------------------------------------------------------------------------------- + TOTAL 0.00003022 0.02169359 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 645632 +BPFP 4.1755 bits/point +EBPFP 7.5224 equivalent bits/point +MSE 0.021694 +---------------------- -------------------------------------------------------- +Time: 1.249s Load: 0.010s, Pack+Encode: 0.418s, Decode+Unpack: 0.821s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0217 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample67-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample67-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample68-layer4-item1.zst (76/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample68-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,552B, BPFP=3.8789 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,540B, BPFP=2.9626 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,048B, BPFP=3.9702 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,540B, BPFP=11.6064 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,812B, BPFP=3.7117 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,164B, BPFP=11.5146 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,072B, BPFP=3.4834 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,936B, BPFP=10.4824 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,308B, BPFP=2.4569 +⌛️ [2/4] FRONTEND: Frontend time: 0.422s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007268 + layer.1.conv_state 0.00050838 0.40284863 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00010671 0.07087411 + layer.3.ssm_state 0.00000001 0.00000473 + layer.3.conv_state 0.00007018 0.06222994 + layer.4.ssm_state 0.00000002 0.00000602 + layer.4.conv_state 0.00026429 0.13719904 + layer.4.output 0.00000311 0.00065325 + ------------------------------------------------------------------------------------- + TOTAL 0.00003009 0.02153417 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 645804 +BPFP 4.1766 bits/point +EBPFP 7.5234 equivalent bits/point +MSE 0.021534 +---------------------- -------------------------------------------------------- +Time: 1.249s Load: 0.009s, Pack+Encode: 0.422s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample68-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample68-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample7-layer4-item1.zst (77/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample7-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 104, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,404B, BPFP=3.8699 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,568B, BPFP=2.9644 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,192B, BPFP=10.3008 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 64,956B, BPFP=3.9646 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,460B, BPFP=11.5869 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,392B, BPFP=3.6860 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,220B, BPFP=11.5283 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,140B, BPFP=3.4875 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,776B, BPFP=10.4434 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,316B, BPFP=2.4098 +⌛️ [2/4] FRONTEND: Frontend time: 0.427s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.826s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007267 + layer.1.conv_state 0.00051536 0.40238425 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00012850 0.07096257 + layer.3.ssm_state 0.00000001 0.00000460 + layer.3.conv_state 0.00007391 0.06256397 + layer.4.ssm_state 0.00000003 0.00000613 + layer.4.conv_state 0.00023144 0.14048442 + layer.4.output 0.00000309 0.00051194 + ------------------------------------------------------------------------------------- + TOTAL 0.00002990 0.02143380 + (elements=1,245,184) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1245184 +Total Bytes 645048 +BPFP 4.1443 bits/point +EBPFP 7.4641 equivalent bits/point +MSE 0.021434 +---------------------- -------------------------------------------------------- +Time: 1.261s Load: 0.008s, Pack+Encode: 0.427s, Decode+Unpack: 0.826s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample7-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample7-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample71-layer4-item1.zst (78/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample71-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,560B, BPFP=3.8794 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,688B, BPFP=2.9717 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,216B, BPFP=10.3066 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,184B, BPFP=3.9785 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,540B, BPFP=11.6064 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,976B, BPFP=3.7217 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,184B, BPFP=11.5195 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,500B, BPFP=3.5095 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,832B, BPFP=10.4570 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,768B, BPFP=2.4228 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.825s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007341 + layer.1.conv_state 0.00051718 0.40359145 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00013311 0.07082162 + layer.3.ssm_state 0.00000001 0.00000457 + layer.3.conv_state 0.00006873 0.06256615 + layer.4.ssm_state 0.00000004 0.00000597 + layer.4.conv_state 0.00022143 0.13547961 + layer.4.output 0.00000284 0.00064530 + ------------------------------------------------------------------------------------- + TOTAL 0.00002967 0.02144431 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 646072 +BPFP 4.1646 bits/point +EBPFP 7.5055 equivalent bits/point +MSE 0.021444 +---------------------- -------------------------------------------------------- +Time: 1.253s Load: 0.009s, Pack+Encode: 0.420s, Decode+Unpack: 0.825s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample71-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample71-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample72-layer4-item1.zst (79/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample72-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,508B, BPFP=3.8762 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,528B, BPFP=2.9619 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,172B, BPFP=10.2959 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,068B, BPFP=3.9714 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,580B, BPFP=11.6162 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,688B, BPFP=3.7041 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,192B, BPFP=11.5215 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,556B, BPFP=3.5129 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,968B, BPFP=10.4902 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,784B, BPFP=2.4231 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.833s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007312 + layer.1.conv_state 0.00050321 0.40227455 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00014768 0.07107913 + layer.3.ssm_state 0.00000001 0.00000480 + layer.3.conv_state 0.00007730 0.06334167 + layer.4.ssm_state 0.00000002 0.00000600 + layer.4.conv_state 0.00022845 0.13518976 + layer.4.output 0.00000298 0.00064265 + ------------------------------------------------------------------------------------- + TOTAL 0.00003014 0.02142826 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 645668 +BPFP 4.1619 bits/point +EBPFP 7.5002 equivalent bits/point +MSE 0.021428 +---------------------- -------------------------------------------------------- +Time: 1.261s Load: 0.009s, Pack+Encode: 0.420s, Decode+Unpack: 0.833s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample72-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample72-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample74-layer4-item1.zst (80/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample74-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 101, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,396B, BPFP=3.8694 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,784B, BPFP=2.9775 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,208B, BPFP=10.3047 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,044B, BPFP=3.9700 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,496B, BPFP=11.5957 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,932B, BPFP=3.7190 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,136B, BPFP=11.5078 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,424B, BPFP=3.5049 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,644B, BPFP=10.4111 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,344B, BPFP=2.4626 +⌛️ [2/4] FRONTEND: Frontend time: 0.424s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.835s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000011 0.00007399 + layer.1.conv_state 0.00050786 0.40295202 + layer.2.ssm_state 0.00000001 0.00000360 + layer.2.conv_state 0.00013194 0.07076182 + layer.3.ssm_state 0.00000001 0.00000469 + layer.3.conv_state 0.00006897 0.06240934 + layer.4.ssm_state 0.00000002 0.00000563 + layer.4.conv_state 0.00028450 0.13064463 + layer.4.output 0.00000302 0.00054222 + ------------------------------------------------------------------------------------- + TOTAL 0.00003131 0.02139670 + (elements=1,232,896) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1232896 +Total Bytes 645032 +BPFP 4.1855 bits/point +EBPFP 7.5446 equivalent bits/point +MSE 0.021397 +---------------------- -------------------------------------------------------- +Time: 1.268s Load: 0.009s, Pack+Encode: 0.424s, Decode+Unpack: 0.835s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample74-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample74-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample75-layer4-item1.zst (81/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample75-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 104, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,564B, BPFP=3.8796 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,540B, BPFP=2.9626 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,228B, BPFP=10.3096 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,104B, BPFP=3.9736 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,608B, BPFP=11.6230 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,632B, BPFP=3.7007 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,188B, BPFP=11.5205 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,352B, BPFP=3.5005 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,836B, BPFP=10.4580 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,420B, BPFP=2.4117 +⌛️ [2/4] FRONTEND: Frontend time: 0.423s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.827s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 104, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000250 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007284 + layer.1.conv_state 0.00051907 0.40363091 + layer.2.ssm_state 0.00000001 0.00000364 + layer.2.conv_state 0.00013541 0.07087991 + layer.3.ssm_state 0.00000001 0.00000463 + layer.3.conv_state 0.00006970 0.06281898 + layer.4.ssm_state 0.00000001 0.00000585 + layer.4.conv_state 0.00021955 0.13581337 + layer.4.output 0.00000291 0.00051423 + ------------------------------------------------------------------------------------- + TOTAL 0.00002969 0.02134900 + (elements=1,245,184) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1245184 +Total Bytes 646096 +BPFP 4.1510 bits/point +EBPFP 7.4769 equivalent bits/point +MSE 0.021349 +---------------------- -------------------------------------------------------- +Time: 1.261s Load: 0.010s, Pack+Encode: 0.423s, Decode+Unpack: 0.827s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0213 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample75-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample75-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample78-layer4-item1.zst (82/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample78-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 101, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,380B, BPFP=3.8684 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,676B, BPFP=2.9709 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,176B, BPFP=10.2969 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,040B, BPFP=3.9697 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,564B, BPFP=11.6123 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,552B, BPFP=3.6958 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,156B, BPFP=11.5127 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,360B, BPFP=3.5010 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,016B, BPFP=10.5020 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 126,008B, BPFP=2.4367 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007308 + layer.1.conv_state 0.00050645 0.40126485 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00013674 0.07074393 + layer.3.ssm_state 0.00000001 0.00000478 + layer.3.conv_state 0.00006915 0.06280728 + layer.4.ssm_state 0.00000001 0.00000590 + layer.4.conv_state 0.00024526 0.13482642 + layer.4.output 0.00000306 0.00052772 + ------------------------------------------------------------------------------------- + TOTAL 0.00003040 0.02146818 + (elements=1,232,896) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1232896 +Total Bytes 643552 +BPFP 4.1759 bits/point +EBPFP 7.5341 equivalent bits/point +MSE 0.021468 +---------------------- -------------------------------------------------------- +Time: 1.246s Load: 0.009s, Pack+Encode: 0.419s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample78-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample78-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample8-layer4-item1.zst (83/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample8-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 105, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,440B, BPFP=3.8721 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,640B, BPFP=2.9688 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,264B, BPFP=10.3184 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,040B, BPFP=3.9697 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,592B, BPFP=11.6191 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,744B, BPFP=3.7075 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,260B, BPFP=11.5381 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,140B, BPFP=3.4875 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,984B, BPFP=10.4941 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 130,376B, BPFP=2.4251 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.816s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007257 + layer.1.conv_state 0.00051340 0.40300393 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00013703 0.07114910 + layer.3.ssm_state 0.00000001 0.00000472 + layer.3.conv_state 0.00007042 0.06361574 + layer.4.ssm_state 0.00000002 0.00000595 + layer.4.conv_state 0.00024895 0.13575688 + layer.4.output 0.00000292 0.00048708 + ------------------------------------------------------------------------------------- + TOTAL 0.00003030 0.02128136 + (elements=1,249,280) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1249280 +Total Bytes 648104 +BPFP 4.1503 bits/point +EBPFP 7.4656 equivalent bits/point +MSE 0.021281 +---------------------- -------------------------------------------------------- +Time: 1.242s Load: 0.009s, Pack+Encode: 0.417s, Decode+Unpack: 0.816s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0213 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample8-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample8-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample80-layer4-item1.zst (84/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample80-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 101, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,536B, BPFP=3.8779 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,700B, BPFP=2.9724 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,196B, BPFP=10.3018 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,100B, BPFP=3.9734 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,560B, BPFP=11.6113 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,520B, BPFP=3.6938 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,156B, BPFP=11.5127 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,528B, BPFP=3.5112 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,792B, BPFP=10.4473 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 124,424B, BPFP=2.4061 +⌛️ [2/4] FRONTEND: Frontend time: 0.416s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.816s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007170 + layer.1.conv_state 0.00050642 0.40258104 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00011949 0.07036679 + layer.3.ssm_state 0.00000001 0.00000466 + layer.3.conv_state 0.00006817 0.06287307 + layer.4.ssm_state 0.00000004 0.00000575 + layer.4.conv_state 0.00024581 0.13411626 + layer.4.output 0.00000308 0.00053320 + ------------------------------------------------------------------------------------- + TOTAL 0.00002991 0.02147767 + (elements=1,232,896) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1232896 +Total Bytes 642136 +BPFP 4.1667 bits/point +EBPFP 7.5260 equivalent bits/point +MSE 0.021478 +---------------------- -------------------------------------------------------- +Time: 1.241s Load: 0.009s, Pack+Encode: 0.416s, Decode+Unpack: 0.816s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample80-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample80-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample81-layer4-item1.zst (85/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample81-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 99, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,540B, BPFP=3.8782 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,464B, BPFP=2.9580 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,192B, BPFP=10.3008 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,068B, BPFP=3.9714 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,552B, BPFP=11.6094 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,596B, BPFP=3.6985 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,264B, BPFP=11.5391 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,064B, BPFP=3.4829 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,968B, BPFP=10.4902 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 126,212B, BPFP=2.4900 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.815s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007307 + layer.1.conv_state 0.00050345 0.40179661 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00011822 0.07087136 + layer.3.ssm_state 0.00000001 0.00000467 + layer.3.conv_state 0.00007003 0.06345484 + layer.4.ssm_state 0.00000003 0.00000620 + layer.4.conv_state 0.00022051 0.13977244 + layer.4.output 0.00000301 0.00066514 + ------------------------------------------------------------------------------------- + TOTAL 0.00002933 0.02182107 + (elements=1,224,704) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1224704 +Total Bytes 643544 +BPFP 4.2038 bits/point +EBPFP 7.5831 equivalent bits/point +MSE 0.021821 +---------------------- -------------------------------------------------------- +Time: 1.240s Load: 0.008s, Pack+Encode: 0.417s, Decode+Unpack: 0.815s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample81-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample81-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample83-layer4-item1.zst (86/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample83-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,484B, BPFP=3.8748 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,396B, BPFP=2.9539 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,148B, BPFP=10.2900 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,060B, BPFP=3.9709 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,468B, BPFP=11.5889 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,664B, BPFP=3.7026 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,224B, BPFP=11.5293 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,556B, BPFP=3.5129 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,996B, BPFP=10.4971 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,364B, BPFP=2.4580 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007259 + layer.1.conv_state 0.00050585 0.40187797 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00012980 0.07052213 + layer.3.ssm_state 0.00000001 0.00000460 + layer.3.conv_state 0.00006838 0.06286237 + layer.4.ssm_state 0.00000001 0.00000585 + layer.4.conv_state 0.00027100 0.13431664 + layer.4.output 0.00000302 0.00063415 + ------------------------------------------------------------------------------------- + TOTAL 0.00003074 0.02143303 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 645984 +BPFP 4.1778 bits/point +EBPFP 7.5254 equivalent bits/point +MSE 0.021433 +---------------------- -------------------------------------------------------- +Time: 1.248s Load: 0.008s, Pack+Encode: 0.420s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample83-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample83-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample84-layer4-item1.zst (87/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample84-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 102, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) -> torch.Size([1, 1, 102, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,524B, BPFP=3.8772 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,508B, BPFP=2.9607 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,244B, BPFP=10.3135 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,044B, BPFP=3.9700 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,516B, BPFP=11.6006 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,492B, BPFP=3.6921 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,348B, BPFP=11.5596 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,552B, BPFP=3.5127 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,976B, BPFP=10.4922 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,564B, BPFP=2.4043 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 102, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000027 0.00007215 + layer.1.conv_state 0.00051229 0.40240350 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00012754 0.07069423 + layer.3.ssm_state 0.00000001 0.00000471 + layer.3.conv_state 0.00007197 0.06280152 + layer.4.ssm_state 0.00000004 0.00000604 + layer.4.conv_state 0.00024522 0.13382269 + layer.4.output 0.00000311 0.00064830 + ------------------------------------------------------------------------------------- + TOTAL 0.00003031 0.02144158 + (elements=1,236,992) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1236992 +Total Bytes 643392 +BPFP 4.1610 bits/point +EBPFP 7.5100 equivalent bits/point +MSE 0.021442 +---------------------- -------------------------------------------------------- +Time: 1.246s Load: 0.008s, Pack+Encode: 0.417s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 102, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample84-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample84-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample85-layer4-item1.zst (88/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample85-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 108, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 108, 4096]) -> torch.Size([1, 1, 108, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,568B, BPFP=3.8799 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,528B, BPFP=2.9619 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,176B, BPFP=10.2969 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,024B, BPFP=3.9688 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,484B, BPFP=11.5928 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,548B, BPFP=3.6956 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,324B, BPFP=11.5537 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,892B, BPFP=3.4724 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,904B, BPFP=10.4746 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 131,992B, BPFP=2.3870 +⌛️ [2/4] FRONTEND: Frontend time: 0.416s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 108, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.835s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 108, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000246 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007338 + layer.1.conv_state 0.00051405 0.40267840 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00011333 0.07063426 + layer.3.ssm_state 0.00000001 0.00000486 + layer.3.conv_state 0.00007471 0.06260068 + layer.4.ssm_state 0.00000002 0.00000619 + layer.4.conv_state 0.00029453 0.13961299 + layer.4.output 0.00000297 0.00050762 + ------------------------------------------------------------------------------------- + TOTAL 0.00003074 0.02113811 + (elements=1,261,568) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1261568 +Total Bytes 649064 +BPFP 4.1159 bits/point +EBPFP 7.3948 equivalent bits/point +MSE 0.021138 +---------------------- -------------------------------------------------------- +Time: 1.258s Load: 0.008s, Pack+Encode: 0.416s, Decode+Unpack: 0.835s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0211 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample85-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample85-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample86-layer4-item1.zst (89/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample86-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,540B, BPFP=3.8782 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,652B, BPFP=2.9695 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,204B, BPFP=10.3037 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,080B, BPFP=3.9722 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,488B, BPFP=11.5938 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,584B, BPFP=3.6978 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,276B, BPFP=11.5420 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,484B, BPFP=3.5085 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,888B, BPFP=10.4707 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,892B, BPFP=2.4198 +⌛️ [2/4] FRONTEND: Frontend time: 0.428s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007271 + layer.1.conv_state 0.00051465 0.40243912 + layer.2.ssm_state 0.00000001 0.00000365 + layer.2.conv_state 0.00011702 0.07111347 + layer.3.ssm_state 0.00000001 0.00000484 + layer.3.conv_state 0.00007124 0.06294073 + layer.4.ssm_state 0.00000002 0.00000600 + layer.4.conv_state 0.00024120 0.13609312 + layer.4.output 0.00000314 0.00054085 + ------------------------------------------------------------------------------------- + TOTAL 0.00003014 0.02162085 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 641712 +BPFP 4.1778 bits/point +EBPFP 7.5490 equivalent bits/point +MSE 0.021621 +---------------------- -------------------------------------------------------- +Time: 1.255s Load: 0.009s, Pack+Encode: 0.428s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0216 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample86-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample86-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample87-layer4-item1.zst (90/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample87-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 101, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) -> torch.Size([1, 1, 101, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,392B, BPFP=3.8691 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,620B, BPFP=2.9675 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,164B, BPFP=10.2939 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,056B, BPFP=3.9707 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,612B, BPFP=11.6240 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,408B, BPFP=3.6870 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,180B, BPFP=11.5186 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,396B, BPFP=3.5032 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,008B, BPFP=10.5000 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,900B, BPFP=2.4346 +⌛️ [2/4] FRONTEND: Frontend time: 0.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 101, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000028 0.00007297 + layer.1.conv_state 0.00050888 0.40119404 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013734 0.07081439 + layer.3.ssm_state 0.00000001 0.00000480 + layer.3.conv_state 0.00006850 0.06276630 + layer.4.ssm_state 0.00000001 0.00000597 + layer.4.conv_state 0.00023156 0.13488565 + layer.4.output 0.00000306 0.00052085 + ------------------------------------------------------------------------------------- + TOTAL 0.00003009 0.02146635 + (elements=1,232,896) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1232896 +Total Bytes 643360 +BPFP 4.1746 bits/point +EBPFP 7.5323 equivalent bits/point +MSE 0.021466 +---------------------- -------------------------------------------------------- +Time: 1.246s Load: 0.009s, Pack+Encode: 0.419s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 101, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample87-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample87-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample88-layer4-item1.zst (91/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample88-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.010s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,632B, BPFP=3.8838 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,544B, BPFP=2.9629 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,092B, BPFP=10.2764 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,200B, BPFP=3.9795 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,512B, BPFP=11.5996 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,648B, BPFP=3.7017 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,100B, BPFP=11.4990 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,980B, BPFP=3.4778 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,872B, BPFP=10.4668 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 127,716B, BPFP=2.4218 +⌛️ [2/4] FRONTEND: Frontend time: 0.439s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.818s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007238 + layer.1.conv_state 0.00050284 0.40187007 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00012269 0.07083028 + layer.3.ssm_state 0.00000001 0.00000476 + layer.3.conv_state 0.00006872 0.06312159 + layer.4.ssm_state 0.00000003 0.00000611 + layer.4.conv_state 0.00023958 0.14008863 + layer.4.output 0.00000284 0.00063715 + ------------------------------------------------------------------------------------- + TOTAL 0.00002950 0.02153260 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 644920 +BPFP 4.1571 bits/point +EBPFP 7.4910 equivalent bits/point +MSE 0.021533 +---------------------- -------------------------------------------------------- +Time: 1.267s Load: 0.010s, Pack+Encode: 0.439s, Decode+Unpack: 0.818s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0215 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample88-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample88-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample9-layer4-item1.zst (92/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample9-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 105, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) -> torch.Size([1, 1, 105, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,508B, BPFP=3.8762 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,528B, BPFP=2.9619 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,212B, BPFP=10.3057 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,060B, BPFP=3.9709 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,568B, BPFP=11.6133 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,660B, BPFP=3.7024 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,212B, BPFP=11.5264 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,996B, BPFP=3.4788 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,944B, BPFP=10.4844 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 130,372B, BPFP=2.4251 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.823s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 105, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007296 + layer.1.conv_state 0.00051499 0.40460849 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00014898 0.07105816 + layer.3.ssm_state 0.00000001 0.00000472 + layer.3.conv_state 0.00006960 0.06338712 + layer.4.ssm_state 0.00000003 0.00000619 + layer.4.conv_state 0.00022645 0.13744324 + layer.4.output 0.00000285 0.00047982 + ------------------------------------------------------------------------------------- + TOTAL 0.00003001 0.02135687 + (elements=1,249,280) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1249280 +Total Bytes 647684 +BPFP 4.1476 bits/point +EBPFP 7.4603 equivalent bits/point +MSE 0.021357 +---------------------- -------------------------------------------------------- +Time: 1.252s Load: 0.009s, Pack+Encode: 0.420s, Decode+Unpack: 0.823s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 105, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample9-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample9-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample91-layer4-item1.zst (93/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample91-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,528B, BPFP=3.8774 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,844B, BPFP=2.9812 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,228B, BPFP=10.3096 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,156B, BPFP=3.9768 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,532B, BPFP=11.6045 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,720B, BPFP=3.7061 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,108B, BPFP=11.5010 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,320B, BPFP=3.4985 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,804B, BPFP=10.4502 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 121,752B, BPFP=2.4515 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.819s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007310 + layer.1.conv_state 0.00051394 0.40151116 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00013358 0.07083227 + layer.3.ssm_state 0.00000001 0.00000463 + layer.3.conv_state 0.00006601 0.06293397 + layer.4.ssm_state 0.00000002 0.00000604 + layer.4.conv_state 0.00022760 0.13851374 + layer.4.output 0.00000310 0.00058521 + ------------------------------------------------------------------------------------- + TOTAL 0.00003032 0.02188073 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 639616 +BPFP 4.2062 bits/point +EBPFP 7.6118 equivalent bits/point +MSE 0.021881 +---------------------- -------------------------------------------------------- +Time: 1.245s Load: 0.009s, Pack+Encode: 0.417s, Decode+Unpack: 0.819s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0219 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample91-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample91-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample92-layer4-item1.zst (94/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample92-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.007s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 103, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,524B, BPFP=3.8772 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,584B, BPFP=2.9653 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,168B, BPFP=10.2949 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,036B, BPFP=3.9695 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,500B, BPFP=11.5967 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,492B, BPFP=3.6921 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,040B, BPFP=11.4844 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,484B, BPFP=3.5085 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,820B, BPFP=10.4541 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 128,456B, BPFP=2.4358 +⌛️ [2/4] FRONTEND: Frontend time: 0.430s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.816s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 103, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007272 + layer.1.conv_state 0.00050709 0.40136486 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00014738 0.07044276 + layer.3.ssm_state 0.00000001 0.00000470 + layer.3.conv_state 0.00006942 0.06292693 + layer.4.ssm_state 0.00000002 0.00000592 + layer.4.conv_state 0.00022996 0.13667417 + layer.4.output 0.00000303 0.00064761 + ------------------------------------------------------------------------------------- + TOTAL 0.00003009 0.02141730 + (elements=1,241,088) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1241088 +Total Bytes 645728 +BPFP 4.1623 bits/point +EBPFP 7.4966 equivalent bits/point +MSE 0.021417 +---------------------- -------------------------------------------------------- +Time: 1.253s Load: 0.007s, Pack+Encode: 0.430s, Decode+Unpack: 0.816s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0214 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample92-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample92-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample93-layer4-item1.zst (95/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample93-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 100, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,568B, BPFP=3.8799 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,756B, BPFP=2.9758 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,144B, BPFP=10.2891 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,188B, BPFP=3.9788 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,492B, BPFP=11.5947 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,404B, BPFP=3.6868 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,128B, BPFP=11.5059 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,328B, BPFP=3.4990 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 43,012B, BPFP=10.5010 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 122,468B, BPFP=2.3920 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.819s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 100, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007237 + layer.1.conv_state 0.00052177 0.40026611 + layer.2.ssm_state 0.00000001 0.00000363 + layer.2.conv_state 0.00012015 0.07017961 + layer.3.ssm_state 0.00000001 0.00000469 + layer.3.conv_state 0.00007075 0.06302933 + layer.4.ssm_state 0.00000003 0.00000625 + layer.4.conv_state 0.00024699 0.13938409 + layer.4.output 0.00000299 0.00053284 + ------------------------------------------------------------------------------------- + TOTAL 0.00003050 0.02162542 + (elements=1,228,800) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1228800 +Total Bytes 640112 +BPFP 4.1674 bits/point +EBPFP 7.5375 equivalent bits/point +MSE 0.021625 +---------------------- -------------------------------------------------------- +Time: 1.244s Load: 0.009s, Pack+Encode: 0.417s, Decode+Unpack: 0.819s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0216 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample93-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample93-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample94-layer4-item1.zst (96/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample94-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 99, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,536B, BPFP=3.8779 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,460B, BPFP=2.9578 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,240B, BPFP=10.3125 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,100B, BPFP=3.9734 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,524B, BPFP=11.6025 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,680B, BPFP=3.7036 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,088B, BPFP=11.4961 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,040B, BPFP=3.4814 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,760B, BPFP=10.4395 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 124,712B, BPFP=2.4604 +⌛️ [2/4] FRONTEND: Frontend time: 0.416s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.814s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000012 0.00007351 + layer.1.conv_state 0.00050618 0.40378588 + layer.2.ssm_state 0.00000001 0.00000357 + layer.2.conv_state 0.00013466 0.07062789 + layer.3.ssm_state 0.00000001 0.00000467 + layer.3.conv_state 0.00006821 0.06264813 + layer.4.ssm_state 0.00000002 0.00000590 + layer.4.conv_state 0.00020718 0.13766924 + layer.4.output 0.00000289 0.00064838 + ------------------------------------------------------------------------------------- + TOTAL 0.00002940 0.02178439 + (elements=1,224,704) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1224704 +Total Bytes 641764 +BPFP 4.1921 bits/point +EBPFP 7.5696 equivalent bits/point +MSE 0.021784 +---------------------- -------------------------------------------------------- +Time: 1.238s Load: 0.008s, Pack+Encode: 0.416s, Decode+Unpack: 0.814s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample94-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample94-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample95-layer4-item1.zst (97/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample95-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 99, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,564B, BPFP=3.8796 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,472B, BPFP=2.9585 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,232B, BPFP=10.3105 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,100B, BPFP=3.9734 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,580B, BPFP=11.6162 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,872B, BPFP=3.7153 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,028B, BPFP=11.4814 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,072B, BPFP=3.4834 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,916B, BPFP=10.4775 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 125,236B, BPFP=2.4707 +⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.810s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000249 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007351 + layer.1.conv_state 0.00050742 0.40405267 + layer.2.ssm_state 0.00000001 0.00000357 + layer.2.conv_state 0.00013289 0.07051583 + layer.3.ssm_state 0.00000001 0.00000468 + layer.3.conv_state 0.00006852 0.06279778 + layer.4.ssm_state 0.00000003 0.00000594 + layer.4.conv_state 0.00023682 0.13813536 + layer.4.output 0.00000300 0.00066215 + ------------------------------------------------------------------------------------- + TOTAL 0.00003022 0.02180956 + (elements=1,224,704) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1224704 +Total Bytes 642696 +BPFP 4.1982 bits/point +EBPFP 7.5784 equivalent bits/point +MSE 0.021810 +---------------------- -------------------------------------------------------- +Time: 1.237s Load: 0.009s, Pack+Encode: 0.417s, Decode+Unpack: 0.810s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample95-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample95-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample96-layer4-item1.zst (98/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample96-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 95, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,368B, BPFP=3.8677 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,724B, BPFP=2.9739 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,136B, BPFP=10.2871 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,088B, BPFP=3.9727 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,532B, BPFP=11.6045 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,532B, BPFP=3.6946 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,088B, BPFP=11.4961 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,152B, BPFP=3.4883 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,832B, BPFP=10.4570 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 121,308B, BPFP=2.4940 +⌛️ [2/4] FRONTEND: Frontend time: 0.423s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 95, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 95, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007265 + layer.1.conv_state 0.00050036 0.40126595 + layer.2.ssm_state 0.00000001 0.00000361 + layer.2.conv_state 0.00012267 0.07123319 + layer.3.ssm_state 0.00000001 0.00000475 + layer.3.conv_state 0.00007107 0.06277525 + layer.4.ssm_state 0.00000002 0.00000614 + layer.4.conv_state 0.00023296 0.13735566 + layer.4.output 0.00000331 0.00064800 + ------------------------------------------------------------------------------------- + TOTAL 0.00003019 0.02201382 + (elements=1,208,320) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1208320 +Total Bytes 638384 +BPFP 4.2266 bits/point +EBPFP 7.6500 equivalent bits/point +MSE 0.022014 +---------------------- -------------------------------------------------------- +Time: 1.251s Load: 0.008s, Pack+Encode: 0.423s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0220 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample96-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample96-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample97-layer4-item1.zst (99/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample97-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.008s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 99, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,360B, BPFP=3.8672 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,772B, BPFP=2.9768 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,216B, BPFP=10.3066 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,080B, BPFP=3.9722 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,664B, BPFP=11.6367 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,728B, BPFP=3.7065 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,152B, BPFP=11.5117 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 57,152B, BPFP=3.4883 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,780B, BPFP=10.4443 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 124,416B, BPFP=2.4545 +⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.820s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 99, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000247 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000010 0.00007313 + layer.1.conv_state 0.00050553 0.40370536 + layer.2.ssm_state 0.00000001 0.00000362 + layer.2.conv_state 0.00013994 0.07085268 + layer.3.ssm_state 0.00000001 0.00000476 + layer.3.conv_state 0.00007011 0.06268708 + layer.4.ssm_state 0.00000002 0.00000596 + layer.4.conv_state 0.00023247 0.13818149 + layer.4.output 0.00000297 0.00066175 + ------------------------------------------------------------------------------------- + TOTAL 0.00003027 0.02180740 + (elements=1,224,704) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1224704 +Total Bytes 641944 +BPFP 4.1933 bits/point +EBPFP 7.5739 equivalent bits/point +MSE 0.021807 +---------------------- -------------------------------------------------------- +Time: 1.245s Load: 0.008s, Pack+Encode: 0.418s, Decode+Unpack: 0.820s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0218 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample97-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample97-layer4-item1.zst + + 💪 Processing: ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample99-layer4-item1.zst (100/100) + +[1/4] FRONTEND: Loading features from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample99-layer4-item1.zst... + +Original data structure: +root: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +⌛️ [1/4] FRONTEND: Load time: 0.009s + +------------------------------------------------------------ +FalconMamba Features Summary +------------------------------------------------------------ +Number of layers: 5 +SSM state shape: (1, 8192, 16) +Conv state shape: (1, 8192, 4) +Output shape: (1, 97, 4096) +Data type: torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) -> torch.Size([1, 1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) -> torch.Size([1, 1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + layer.0.ssm_state: 63,396B, BPFP=3.8694 + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + layer.0.conv_state: 42,624B, BPFP=10.4062 + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + layer.1.ssm_state: 48,560B, BPFP=2.9639 + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + layer.1.conv_state: 42,224B, BPFP=10.3086 + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + layer.2.ssm_state: 65,072B, BPFP=3.9717 + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + layer.2.conv_state: 47,616B, BPFP=11.6250 + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + layer.3.ssm_state: 60,708B, BPFP=3.7053 + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + layer.3.conv_state: 47,244B, BPFP=11.5342 + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + layer.4.ssm_state: 56,760B, BPFP=3.4644 + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + layer.4.conv_state: 42,868B, BPFP=10.4658 + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + layer.4.output: 123,596B, BPFP=2.4886 +⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) +⌛️ [3/4] BACKEND: Backend time: 0.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.0.ssm_state: torch.Size([256])) for layer.0.ssm_state + Using per-key quantization points (layer.0.conv_state: torch.Size([256])) for layer.0.conv_state + Using per-key quantization points (layer.1.ssm_state: torch.Size([256])) for layer.1.ssm_state + Using per-key quantization points (layer.1.conv_state: torch.Size([256])) for layer.1.conv_state + Using per-key quantization points (layer.2.ssm_state: torch.Size([256])) for layer.2.ssm_state + Using per-key quantization points (layer.2.conv_state: torch.Size([256])) for layer.2.conv_state + Using per-key quantization points (layer.3.ssm_state: torch.Size([256])) for layer.3.ssm_state + Using per-key quantization points (layer.3.conv_state: torch.Size([256])) for layer.3.conv_state + Using per-key quantization points (layer.4.ssm_state: torch.Size([256])) for layer.4.ssm_state + Using per-key quantization points (layer.4.conv_state: torch.Size([256])) for layer.4.conv_state + Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output + IndividualUnPacker: + layer.0.ssm_state: torch.Size([1, 8192, 16]) + layer.0.conv_state: torch.Size([1, 8192, 4]) + layer.1.ssm_state: torch.Size([1, 8192, 16]) + layer.1.conv_state: torch.Size([1, 8192, 4]) + layer.2.ssm_state: torch.Size([1, 8192, 16]) + layer.2.conv_state: torch.Size([1, 8192, 4]) + layer.3.ssm_state: torch.Size([1, 8192, 16]) + layer.3.conv_state: torch.Size([1, 8192, 4]) + layer.4.ssm_state: torch.Size([1, 8192, 16]) + layer.4.conv_state: torch.Size([1, 8192, 4]) + layer.4.output: torch.Size([1, 97, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.0.ssm_state 0.00000001 0.00000248 + layer.0.conv_state 0.00014612 0.13107595 + layer.1.ssm_state 0.00000013 0.00007338 + layer.1.conv_state 0.00050423 0.40512708 + layer.2.ssm_state 0.00000001 0.00000359 + layer.2.conv_state 0.00013920 0.07116281 + layer.3.ssm_state 0.00000001 0.00000471 + layer.3.conv_state 0.00007129 0.06331539 + layer.4.ssm_state 0.00000002 0.00000624 + layer.4.conv_state 0.00024943 0.14073132 + layer.4.output 0.00000302 0.00059154 + ------------------------------------------------------------------------------------- + TOTAL 0.00003091 0.02205917 + (elements=1,216,512) +---------------------- -------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- -------------------------------------------------------- +Handler falconmamba +Strategy individual +Architecture hyperprior-featurecoding +---------------------- -------------------------------------------------------- +Total Elements 1216512 +Total Bytes 640668 +BPFP 4.2131 bits/point +EBPFP 7.6135 equivalent bits/point +MSE 0.022059 +---------------------- -------------------------------------------------------- +Time: 1.251s Load: 0.009s, Pack+Encode: 0.420s, Decode+Unpack: 0.822s +---------------------- -------------------------------------------------------- +Restored Feature Format: [dict] with 3 keys + key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.float32, device=cpu + key['ssm_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 16]), dtype=torch.bfloat16, device=cpu + key['conv_state']: [list] with 5 items + item[0]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[1]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[2]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[3]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu + item[4]: [Tensor] shape=torch.Size([1, 8192, 4]), dtype=torch.bfloat16, device=cpu +💾 Converting with 0.0221 MSE: + from ../datasets/FalconMamba-500features-L5wCache/tiiuae/falcon-mamba-7b-instruct/fc_winogrande/sample99-layer4-item1.zst + to output-fixed/falconmamba/lambda0.02/hyperprior-featurecoding-8bit-individual/fc_winogrande/sample99-layer4-item1.zst +------------------------ ---------------------------- +TOTAL PROCESSING SUMMARY +------------------------ ---------------------------- +Total files 100 +Avg BPFP 4.1712 bits/point +Avg EBPFP 7.5232 equivalent bits/point +Avg MSE 0.021534 +Avg Time 1.263s +------------------------ ----------------------------