diff --git a/.gitattributes b/.gitattributes index 26bb61d0b722cea3038bac096f443aaa4de2b8ad..2789e57d45ef79ba0323c483000431f5fd32e863 100644 --- a/.gitattributes +++ b/.gitattributes @@ -5091,3 +5091,4 @@ lambda0.001/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample35-layer4 lambda0.001/elic-featurecoding-8bit-individual/fc_truthfulqa_mc1/sample43-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample155-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text lambda0.001/elic-featurecoding-8bit-individual/fc_winogrande/sample47-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample454-layer4-item1.zst filter=lfs diff=lfs merge=lfs -text diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample0-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample0-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..28c1b16cb4e8a1d7993ff43cffbd546de59b1838 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample0-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3e42cfc88ebfd693dd783efbf3a1e6848ddd0c65b8d68d32365f7fd0210b73d +size 3228133 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample1-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample1-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..faf77641c7fccf22a4898a0c9eb2684d6491f557 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample1-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:433376c9d95d8e407fa8af5ca2dda7224cd10fd3ee301bd6a4f0f3e5c4fa929e +size 3487458 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample10-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample10-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..70d7d62e66c10942c6124e9774676d09032921de --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample10-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b12b27076cde44bbb9aa53c6949aab99125d654caf471d893cb8b4c7f0ef36b2 +size 2893933 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample100-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample100-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..8a2696012858640ec2451f37c34d500c3f70c323 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample100-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9acdee4dacc92a70ea37c4a9155d8c084d67289a714a4e9d865774253a2c21d +size 2303218 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample101-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample101-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..2319aa4b7f7a59f79cb0733630e610e19091c0f0 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample101-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b40ffaaf337a8381c6f44360a0c66f3a52059f5ccfb4e73313cdd0d4ddab109 +size 2127780 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample102-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample102-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..8992e7f1b08dc8cbeee86f5ce2813150daef891e --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample102-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b1ef040905b42510ae055df59ffd683de5884bbe90a35ac551fccbda66af6c8 +size 2054445 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample103-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample103-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..4ea4e2ebf90773b92a47efa68f53d99532c06434 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample103-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd2ddc23e8b0c0ddb790bde5783ab30472092e5d29996fddfbeab8856542307b +size 2044046 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample104-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample104-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..977b4bcfd0327c66487f2f1aa52b1b742eb204de --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample104-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ab456480586262fd37fb6bc259be5ea1608ff500bf561ae1314c4ac5864ae8a +size 2446779 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample105-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample105-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..c28a870d6e57a60267519499adac410add3f7860 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample105-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03824fb1489b16ae1bba9e4c461cfc8bfcde34ab36dc5b0f56c012558684d188 +size 2342820 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample106-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample106-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..fc690f42e5a8fd624e715951edeb1db647da330a --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample106-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62d44184ffb62c2065a896d0151756f63b54c12e354c8e0dcbb3ce9fc8e3dfe1 +size 2583962 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample107-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample107-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..2dfbbe43204973766af061c526217e0dd80400f6 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample107-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d3aa97bc46bb305d42cc2b0268e1107ed82e9654e724490310076840f30564a +size 2105985 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample108-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample108-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..e9db0fd40727323671236502f818da4b69a1c939 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample108-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:beb408b579c8ce599e0b1f8ab6ffacaf08aff62c8a46347dfdadaf044c59e02d +size 2099784 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample109-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample109-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..a1bedefb452978a7ba08e5b59272dbf763ccac7a --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample109-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e918c0f8f59b3614768fd06f75f09c540720790cabe1168acd7da95d4ad1584 +size 2352939 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample11-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample11-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..ac3a617438c13d9321d305257d6d5e3d5748a0b7 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample11-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d41103d4a148b886be97d3f4cf9cf343920250c3c1ba30d69197c93aacb5872 +size 2594814 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample111-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample111-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..c671d4b590db7a83595d2eca393790bb5d0adbe9 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample111-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:927af56965d450bdeaae204a177bb4c530b426c4852beedbd91d519bfb412772 +size 2170704 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample112-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample112-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..80927badbe69396038b8647abf287f4f4280446a --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample112-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:002cadd7c1451bfb80e3ebc34eaf2610b78fd1557d84b180edad875bf782a5e3 +size 2105541 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample12-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample12-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..5506798f45b05ebbd399b9af570a051155416126 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample12-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13496e2afe978ba43c4cc560237045f12c9af4d2e262cfbdba8f4bb565feb701 +size 2640101 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample13-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample13-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..3f33b6849e2c53d950e8e20e1e1afc8f55092e10 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample13-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09876cc5f8c16ea6348f8f6a3192145c391f9692a66cf5b5a0570ecbeb7c4263 +size 2700003 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample14-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample14-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..a2ddbf3bf084a14e569c0da2b9f653f4fd977257 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample14-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:664804230d0f6e69c1acd2ac84a793210ae2e05dfa46da7a93ca1eae50fe3c64 +size 2511082 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample16-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample16-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..0d2407b6d23df1ff742db6b3a1fe3c9f36f79466 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample16-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dbe0eda50fb9ffdffb9b46372bd2baa72a18f8743c9be4a706afe621bca4ac6e +size 2733419 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample17-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample17-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..6cb78c488e0d1d292e20e0a70bdc91d6b6108b7b --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample17-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad6b8ff8b3b7b60753c8617a6cb736dc8d1b7abcb0a4362a9c87d3e110be7630 +size 2802072 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample18-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample18-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..96a0fe0af3d3d5b354b7d70f8ac0424e9641a788 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample18-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04437267ecf3d82da2bc545c4097cb491341ef8d7247f3d3435d13233c3263dd +size 2557988 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample19-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample19-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..7f23cefd5677a6e306c540167ea37f2727b934bc --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample19-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3752f1fbd7484ad3a8c367f3e1d964a58a42b1560d5acf331fcaa8902a6a558a +size 2504752 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample2-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample2-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..75ca28c415bb7956bf21f6b902edabe1742755fe --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample2-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6cc0e21e894d23f5b5ff3dab17f3b85e127ef133d6a33c366e631db0a9f62d3 +size 3180197 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample20-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample20-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..46dc5efdebbf7447a6478a1c7534bab090064ec2 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample20-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae3a81898b380cf101b6fc26cadac4e493abcbfb369e1eb63fd5021e88cf0c52 +size 2518164 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample21-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample21-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..511b57b1e7c3bf28e673015cee3289a314fc7ba5 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample21-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39cc783a84c2e3803c9038fec9bd392a45058c4b5d114b3a88aac51505e8aa97 +size 2400936 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample22-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample22-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..cd1029d5ab4d46f89f21d212aaddac3cab288dd6 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample22-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbad044d4200cb7a7f995ba7020175fff829982e2b889f70c5f13064e093e155 +size 2592751 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample23-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample23-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..1e800132c6a2a882e03302cae54aa803593e152f --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample23-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3a9561e57eb17b9bf62a43c49215d68aa6c7ed458d0253bb9c67342360c361d +size 2493741 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample24-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample24-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..7414092180ed73e4a6f95fbcff69ef72800dbf7c --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample24-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef522f20edd7dd23d1bee396f9fe03086e0e8cd0cb5437881e4f3d8e0c054cc7 +size 2485163 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample25-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample25-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..292b66beddffd1dd9d3a4a947db2c207c43dabbc --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample25-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2ee10e78f90357d9fe2da4e36bb96c3609af799cf83457d27f80bf2a8da8561 +size 2578710 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample26-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample26-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..31c195dbd095f36563117666cf4a46c834411f5c --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample26-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4bb92a2f334e75bbbf2d5635f04af766a376823a8b8173496632d7bd2657c95a +size 2598965 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample27-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample27-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..1da8813a5998cecef02de69f0c0bc19b13cf1bbf --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample27-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d55929679d771454362a75952171f1aa9c038b0bac0b1321e2544517edda423d +size 2341219 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample28-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample28-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..18997f3794afcd1264bfce7af22d3f16c1fbdcf4 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample28-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f6ec373e29854b7f6055ec114bd215860ab78edfaf9f1142031ef3b0fcbf30d +size 2429449 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample29-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample29-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..ecf6ddc2639c66068235b22d3750b64fabc77cea --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample29-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6dd78fe1bcd3129d4eb9f867a4beff6df457875d57d9ca503f577edb62073df8 +size 2536128 diff --git a/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample30-layer4-item1.zst b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample30-layer4-item1.zst new file mode 100644 index 0000000000000000000000000000000000000000..cba116856d9a7d61d1517fbb9abdc44b8bde2418 --- /dev/null +++ b/lambda0.001/elic-featurecoding-8bit-individual/fc_arc_challenge/sample30-layer4-item1.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:68cf52335a592b5716972154c1c894b4ac878d422cfdcf5ba5125beddeb09ec0 +size 2392134 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log b/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log index 3adbf2c782f93782859286878b6bcc67dd07203e..cafbcc9707f644f8df25fb28ffb65e7e3574df79 100644 --- a/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log @@ -1,16958 +1,3 @@ -Experiment: dtufc_hyperprior-featurecoding_qwen_individual -Log file: output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/dtufc_hyperprior-featurecoding_qwen_individual.log -DTUFCCodecConfig: - arch: hyperprior-featurecoding - handler: qwen - checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.001_epochs600_lr0.0001_bs360_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_lambda0.001_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar -Checkpoint epoch: 405 -Loaded hyperprior-featurecoding (1-channel) on cuda:0 -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.output' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json -Loaded per-key mappings: model=qwen - Keys: ['layer.0.k_cache', 'layer.0.v_cache', 'layer.1.k_cache', 'layer.1.v_cache', 'layer.2.k_cache', 'layer.2.v_cache', 'layer.3.k_cache', 'layer.3.v_cache', 'layer.4.k_cache', 'layer.4.v_cache', 'layer.4.output'] ----------------- ------------------------------------------------------------------------------------------------------------------------------ -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding -Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.001_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar -Transform type kmeans -Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json -Input ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k -Output output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k ----------------- ------------------------------------------------------------------------------------------------------------------------------ -Files found: 100 ----------------------------------------------------------------------- - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1-layer4-item1.zst (1/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 173, 128) -Output shape: (1, 173, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) -> torch.Size([1, 1, 173, 1024]) - layer.4.output: torch.Size([1, 173, 4096]) -> torch.Size([1, 1, 173, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,648B, BPFP=0.0744 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,668B, BPFP=0.0753 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,180B, BPFP=0.0984 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,844B, BPFP=0.0833 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,292B, BPFP=0.1035 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,960B, BPFP=0.0885 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,384B, BPFP=0.1077 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,068B, BPFP=0.0934 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,124B, BPFP=0.0959 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,024B, BPFP=0.0914 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 5,876B, BPFP=0.0663 -⌛️ [2/4] FRONTEND: Frontend time: 0.604s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.422s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 173, 128]) - layer.0.v_cache: torch.Size([1, 8, 173, 128]) - layer.1.k_cache: torch.Size([1, 8, 173, 128]) - layer.1.v_cache: torch.Size([1, 8, 173, 128]) - layer.2.k_cache: torch.Size([1, 8, 173, 128]) - layer.2.v_cache: torch.Size([1, 8, 173, 128]) - layer.3.k_cache: torch.Size([1, 8, 173, 128]) - layer.3.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.k_cache: torch.Size([1, 8, 173, 128]) - layer.4.v_cache: torch.Size([1, 8, 173, 128]) - layer.4.output: torch.Size([1, 173, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02440321 72.08751806 - layer.0.v_cache 0.00000027 0.00065934 - layer.1.k_cache 0.00314874 6.37707237 - layer.1.v_cache 0.00000072 0.00251388 - layer.2.k_cache 0.00114712 2.14522204 - layer.2.v_cache 0.00000110 0.00402859 - layer.3.k_cache 0.00139354 2.57161699 - layer.3.v_cache 0.00000204 0.00689100 - layer.4.k_cache 0.00353492 6.77232123 - layer.4.v_cache 0.00000301 0.01126529 - layer.4.output 0.00018661 0.22234219 - ------------------------------------------------------------------------------------- - TOTAL 0.00245579 6.49060554 - (elements=2,480,128) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2480128 -Total Bytes 26068 -BPFP 0.0841 bits/point -EBPFP 0.1682 equivalent bits/point -MSE 6.490606 ----------------------- -------------------------------------------------------- -Time: 1.036s Load: 0.009s, Pack+Encode: 0.604s, Decode+Unpack: 0.422s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 173, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 173, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.4906 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample1-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample100-layer4-item1.zst (2/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample100-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 109, 128) -Output shape: (1, 109, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.output: torch.Size([1, 109, 4096]) -> torch.Size([1, 1, 109, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,092B, BPFP=0.0783 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,168B, BPFP=0.0837 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,524B, BPFP=0.1092 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,384B, BPFP=0.0992 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,556B, BPFP=0.1115 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,444B, BPFP=0.1035 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,596B, BPFP=0.1144 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,528B, BPFP=0.1095 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,444B, BPFP=0.1035 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,448B, BPFP=0.1038 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,988B, BPFP=0.0715 -⌛️ [2/4] FRONTEND: Frontend time: 0.264s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.output: torch.Size([1, 109, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.293s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.output: torch.Size([1, 109, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02465495 99.98307590 - layer.0.v_cache 0.00000027 0.00068479 - layer.1.k_cache 0.00330346 6.88853413 - layer.1.v_cache 0.00000080 0.00263966 - layer.2.k_cache 0.00115767 2.21300290 - layer.2.v_cache 0.00000114 0.00413382 - layer.3.k_cache 0.00132842 2.72542243 - layer.3.v_cache 0.00000211 0.00708743 - layer.4.k_cache 0.00335301 6.84434635 - layer.4.v_cache 0.00000290 0.01113919 - layer.4.output 0.00024947 0.24698835 - ------------------------------------------------------------------------------------- - TOTAL 0.00248590 8.54771571 - (elements=1,562,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1562624 -Total Bytes 18172 -BPFP 0.0930 bits/point -EBPFP 0.1861 equivalent bits/point -MSE 8.547716 ----------------------- -------------------------------------------------------- -Time: 0.564s Load: 0.006s, Pack+Encode: 0.264s, Decode+Unpack: 0.293s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.5477 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample100-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample100-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample108-layer4-item1.zst (3/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample108-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 956B, BPFP=0.0770 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,096B, BPFP=0.0883 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,404B, BPFP=0.1131 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,284B, BPFP=0.1034 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,544B, BPFP=0.1244 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,396B, BPFP=0.1124 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,520B, BPFP=0.1224 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,412B, BPFP=0.1137 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,368B, BPFP=0.1102 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,412B, BPFP=0.1137 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,660B, BPFP=0.0737 -⌛️ [2/4] FRONTEND: Frontend time: 0.206s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02579871 127.92062661 - layer.0.v_cache 0.00000028 0.00069287 - layer.1.k_cache 0.00331373 6.75351676 - layer.1.v_cache 0.00000086 0.00277570 - layer.2.k_cache 0.00113316 2.25721442 - layer.2.v_cache 0.00000115 0.00429312 - layer.3.k_cache 0.00137718 2.74779833 - layer.3.v_cache 0.00000224 0.00727653 - layer.4.k_cache 0.00334605 6.80071959 - layer.4.v_cache 0.00000324 0.01210867 - layer.4.output 0.00031880 0.25305732 - ------------------------------------------------------------------------------------- - TOTAL 0.00258942 10.53708942 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 17052 -BPFP 0.0981 bits/point -EBPFP 0.1962 equivalent bits/point -MSE 10.537089 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.007s, Pack+Encode: 0.206s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.5371 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample108-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample108-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1087-layer4-item1.zst (4/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1087-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 50, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 50, 128) -Output shape: (1, 50, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.0.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.1.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.1.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.2.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.2.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.3.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.3.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.4.k_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.4.v_cache: torch.Size([1, 8, 50, 128]) -> torch.Size([1, 1, 50, 1024]) - layer.4.output: torch.Size([1, 50, 4096]) -> torch.Size([1, 1, 50, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 584B, BPFP=0.0912 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 608B, BPFP=0.0950 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 808B, BPFP=0.1263 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 692B, BPFP=0.1081 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 844B, BPFP=0.1319 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 752B, BPFP=0.1175 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 872B, BPFP=0.1363 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 784B, BPFP=0.1225 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 752B, BPFP=0.1175 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 736B, BPFP=0.1150 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 2,332B, BPFP=0.0911 -⌛️ [2/4] FRONTEND: Frontend time: 0.215s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 50, 128]) - layer.0.v_cache: torch.Size([1, 8, 50, 128]) - layer.1.k_cache: torch.Size([1, 8, 50, 128]) - layer.1.v_cache: torch.Size([1, 8, 50, 128]) - layer.2.k_cache: torch.Size([1, 8, 50, 128]) - layer.2.v_cache: torch.Size([1, 8, 50, 128]) - layer.3.k_cache: torch.Size([1, 8, 50, 128]) - layer.3.v_cache: torch.Size([1, 8, 50, 128]) - layer.4.k_cache: torch.Size([1, 8, 50, 128]) - layer.4.v_cache: torch.Size([1, 8, 50, 128]) - layer.4.output: torch.Size([1, 50, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.203s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 50, 128]) - layer.0.v_cache: torch.Size([1, 8, 50, 128]) - layer.1.k_cache: torch.Size([1, 8, 50, 128]) - layer.1.v_cache: torch.Size([1, 8, 50, 128]) - layer.2.k_cache: torch.Size([1, 8, 50, 128]) - layer.2.v_cache: torch.Size([1, 8, 50, 128]) - layer.3.k_cache: torch.Size([1, 8, 50, 128]) - layer.3.v_cache: torch.Size([1, 8, 50, 128]) - layer.4.k_cache: torch.Size([1, 8, 50, 128]) - layer.4.v_cache: torch.Size([1, 8, 50, 128]) - layer.4.output: torch.Size([1, 50, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02911560 129.41376953 - layer.0.v_cache 0.00000029 0.00068011 - layer.1.k_cache 0.00403332 7.81498047 - layer.1.v_cache 0.00000087 0.00268241 - layer.2.k_cache 0.00117108 2.50297791 - layer.2.v_cache 0.00000108 0.00413947 - layer.3.k_cache 0.00141424 3.01225098 - layer.3.v_cache 0.00000205 0.00690661 - layer.4.k_cache 0.00319313 6.74711670 - layer.4.v_cache 0.00000283 0.01064251 - layer.4.output 0.00027592 0.32249886 - ------------------------------------------------------------------------------------- - TOTAL 0.00285987 10.77186729 - (elements=716,800) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 716800 -Total Bytes 9764 -BPFP 0.1090 bits/point -EBPFP 0.2179 equivalent bits/point -MSE 10.771867 ----------------------- -------------------------------------------------------- -Time: 0.422s Load: 0.005s, Pack+Encode: 0.215s, Decode+Unpack: 0.203s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 50, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 50, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.7719 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1087-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample1087-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1128-layer4-item1.zst (5/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1128-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 57, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 57, 128) -Output shape: (1, 57, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.0.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.1.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.1.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.2.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.2.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.3.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.3.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.4.k_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.4.v_cache: torch.Size([1, 8, 57, 128]) -> torch.Size([1, 1, 57, 1024]) - layer.4.output: torch.Size([1, 57, 4096]) -> torch.Size([1, 1, 57, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 572B, BPFP=0.0784 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 580B, BPFP=0.0795 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 824B, BPFP=0.1129 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 656B, BPFP=0.0899 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 864B, BPFP=0.1184 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 752B, BPFP=0.1031 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 896B, BPFP=0.1228 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 800B, BPFP=0.1096 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 784B, BPFP=0.1075 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 760B, BPFP=0.1042 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 2,344B, BPFP=0.0803 -⌛️ [2/4] FRONTEND: Frontend time: 0.151s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 57, 128]) - layer.0.v_cache: torch.Size([1, 8, 57, 128]) - layer.1.k_cache: torch.Size([1, 8, 57, 128]) - layer.1.v_cache: torch.Size([1, 8, 57, 128]) - layer.2.k_cache: torch.Size([1, 8, 57, 128]) - layer.2.v_cache: torch.Size([1, 8, 57, 128]) - layer.3.k_cache: torch.Size([1, 8, 57, 128]) - layer.3.v_cache: torch.Size([1, 8, 57, 128]) - layer.4.k_cache: torch.Size([1, 8, 57, 128]) - layer.4.v_cache: torch.Size([1, 8, 57, 128]) - layer.4.output: torch.Size([1, 57, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.200s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 57, 128]) - layer.0.v_cache: torch.Size([1, 8, 57, 128]) - layer.1.k_cache: torch.Size([1, 8, 57, 128]) - layer.1.v_cache: torch.Size([1, 8, 57, 128]) - layer.2.k_cache: torch.Size([1, 8, 57, 128]) - layer.2.v_cache: torch.Size([1, 8, 57, 128]) - layer.3.k_cache: torch.Size([1, 8, 57, 128]) - layer.3.v_cache: torch.Size([1, 8, 57, 128]) - layer.4.k_cache: torch.Size([1, 8, 57, 128]) - layer.4.v_cache: torch.Size([1, 8, 57, 128]) - layer.4.output: torch.Size([1, 57, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02901734 327.22546601 - layer.0.v_cache 0.00000028 0.00074758 - layer.1.k_cache 0.00371865 9.29703990 - layer.1.v_cache 0.00000081 0.00289070 - layer.2.k_cache 0.00114644 2.63368198 - layer.2.v_cache 0.00000109 0.00444242 - layer.3.k_cache 0.00141512 3.23173282 - layer.3.v_cache 0.00000208 0.00778262 - layer.4.k_cache 0.00325023 8.79138237 - layer.4.v_cache 0.00000285 0.01186436 - layer.4.output 0.00023785 0.33488615 - ------------------------------------------------------------------------------------- - TOTAL 0.00282188 25.18189824 - (elements=817,152) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 817152 -Total Bytes 9832 -BPFP 0.0963 bits/point -EBPFP 0.1925 equivalent bits/point -MSE 25.181898 ----------------------- -------------------------------------------------------- -Time: 0.354s Load: 0.004s, Pack+Encode: 0.151s, Decode+Unpack: 0.200s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 57, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 57, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 25.1819 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1128-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample1128-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample117-layer4-item1.zst (6/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample117-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 113, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 113, 128) -Output shape: (1, 113, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.output: torch.Size([1, 113, 4096]) -> torch.Size([1, 1, 113, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,108B, BPFP=0.0766 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,168B, BPFP=0.0808 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,540B, BPFP=0.1065 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,404B, BPFP=0.0971 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,576B, BPFP=0.1090 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,528B, BPFP=0.1056 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,648B, BPFP=0.1139 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,544B, BPFP=0.1067 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,468B, BPFP=0.1015 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,500B, BPFP=0.1037 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,912B, BPFP=0.0676 -⌛️ [2/4] FRONTEND: Frontend time: 0.208s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.output: torch.Size([1, 113, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.output: torch.Size([1, 113, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.03016301 84.54063537 - layer.0.v_cache 0.00000028 0.00068867 - layer.1.k_cache 0.00339028 6.42055707 - layer.1.v_cache 0.00000081 0.00267979 - layer.2.k_cache 0.00113693 2.21509761 - layer.2.v_cache 0.00000114 0.00420456 - layer.3.k_cache 0.00135151 2.68741493 - layer.3.v_cache 0.00000219 0.00735703 - layer.4.k_cache 0.00328814 6.41457887 - layer.4.v_cache 0.00000314 0.01194466 - layer.4.output 0.00017637 0.23230649 - ------------------------------------------------------------------------------------- - TOTAL 0.00286021 7.37388461 - (elements=1,619,968) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1619968 -Total Bytes 18396 -BPFP 0.0908 bits/point -EBPFP 0.1817 equivalent bits/point -MSE 7.373885 ----------------------- -------------------------------------------------------- -Time: 0.496s Load: 0.007s, Pack+Encode: 0.208s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 113, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.3739 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample117-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample117-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample120-layer4-item1.zst (7/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample120-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 108, 128) -Output shape: (1, 108, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.0.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.1.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.1.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.2.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.2.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.3.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.3.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.4.k_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.4.v_cache: torch.Size([1, 8, 108, 128]) -> torch.Size([1, 1, 108, 1024]) - layer.4.output: torch.Size([1, 108, 4096]) -> torch.Size([1, 1, 108, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,072B, BPFP=0.0775 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,160B, BPFP=0.0839 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,544B, BPFP=0.1117 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,288B, BPFP=0.0932 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,572B, BPFP=0.1137 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,400B, BPFP=0.1013 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,624B, BPFP=0.1175 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,472B, BPFP=0.1065 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,468B, BPFP=0.1062 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,452B, BPFP=0.1050 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,020B, BPFP=0.0727 -⌛️ [2/4] FRONTEND: Frontend time: 0.206s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 108, 128]) - layer.0.v_cache: torch.Size([1, 8, 108, 128]) - layer.1.k_cache: torch.Size([1, 8, 108, 128]) - layer.1.v_cache: torch.Size([1, 8, 108, 128]) - layer.2.k_cache: torch.Size([1, 8, 108, 128]) - layer.2.v_cache: torch.Size([1, 8, 108, 128]) - layer.3.k_cache: torch.Size([1, 8, 108, 128]) - layer.3.v_cache: torch.Size([1, 8, 108, 128]) - layer.4.k_cache: torch.Size([1, 8, 108, 128]) - layer.4.v_cache: torch.Size([1, 8, 108, 128]) - layer.4.output: torch.Size([1, 108, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 108, 128]) - layer.0.v_cache: torch.Size([1, 8, 108, 128]) - layer.1.k_cache: torch.Size([1, 8, 108, 128]) - layer.1.v_cache: torch.Size([1, 8, 108, 128]) - layer.2.k_cache: torch.Size([1, 8, 108, 128]) - layer.2.v_cache: torch.Size([1, 8, 108, 128]) - layer.3.k_cache: torch.Size([1, 8, 108, 128]) - layer.3.v_cache: torch.Size([1, 8, 108, 128]) - layer.4.k_cache: torch.Size([1, 8, 108, 128]) - layer.4.v_cache: torch.Size([1, 8, 108, 128]) - layer.4.output: torch.Size([1, 108, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02569981 83.15166558 - layer.0.v_cache 0.00000028 0.00064966 - layer.1.k_cache 0.00351470 6.56636952 - layer.1.v_cache 0.00000073 0.00253694 - layer.2.k_cache 0.00116483 2.23549228 - layer.2.v_cache 0.00000104 0.00393038 - layer.3.k_cache 0.00137767 2.68910500 - layer.3.v_cache 0.00000202 0.00687828 - layer.4.k_cache 0.00329212 6.51280495 - layer.4.v_cache 0.00000308 0.01126872 - layer.4.output 0.00021651 0.24276098 - ------------------------------------------------------------------------------------- - TOTAL 0.00256588 7.29655323 - (elements=1,548,288) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1548288 -Total Bytes 18072 -BPFP 0.0934 bits/point -EBPFP 0.1868 equivalent bits/point -MSE 7.296553 ----------------------- -------------------------------------------------------- -Time: 0.496s Load: 0.007s, Pack+Encode: 0.206s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 108, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 108, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.2966 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample120-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample120-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1295-layer4-item1.zst (8/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1295-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 47, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 47, 128) -Output shape: (1, 47, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.0.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.1.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.1.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.2.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.2.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.3.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.3.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.4.k_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.4.v_cache: torch.Size([1, 8, 47, 128]) -> torch.Size([1, 1, 47, 1024]) - layer.4.output: torch.Size([1, 47, 4096]) -> torch.Size([1, 1, 47, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 592B, BPFP=0.0984 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 640B, BPFP=0.1064 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 824B, BPFP=0.1370 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 684B, BPFP=0.1137 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 856B, BPFP=0.1423 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 736B, BPFP=0.1223 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 876B, BPFP=0.1456 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 800B, BPFP=0.1330 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 764B, BPFP=0.1270 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 752B, BPFP=0.1250 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 2,528B, BPFP=0.1051 -⌛️ [2/4] FRONTEND: Frontend time: 0.155s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 47, 128]) - layer.0.v_cache: torch.Size([1, 8, 47, 128]) - layer.1.k_cache: torch.Size([1, 8, 47, 128]) - layer.1.v_cache: torch.Size([1, 8, 47, 128]) - layer.2.k_cache: torch.Size([1, 8, 47, 128]) - layer.2.v_cache: torch.Size([1, 8, 47, 128]) - layer.3.k_cache: torch.Size([1, 8, 47, 128]) - layer.3.v_cache: torch.Size([1, 8, 47, 128]) - layer.4.k_cache: torch.Size([1, 8, 47, 128]) - layer.4.v_cache: torch.Size([1, 8, 47, 128]) - layer.4.output: torch.Size([1, 47, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.197s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 47, 128]) - layer.0.v_cache: torch.Size([1, 8, 47, 128]) - layer.1.k_cache: torch.Size([1, 8, 47, 128]) - layer.1.v_cache: torch.Size([1, 8, 47, 128]) - layer.2.k_cache: torch.Size([1, 8, 47, 128]) - layer.2.v_cache: torch.Size([1, 8, 47, 128]) - layer.3.k_cache: torch.Size([1, 8, 47, 128]) - layer.3.v_cache: torch.Size([1, 8, 47, 128]) - layer.4.k_cache: torch.Size([1, 8, 47, 128]) - layer.4.v_cache: torch.Size([1, 8, 47, 128]) - layer.4.output: torch.Size([1, 47, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02987218 208.42952128 - layer.0.v_cache 0.00000031 0.00071588 - layer.1.k_cache 0.00397950 8.30864829 - layer.1.v_cache 0.00000076 0.00260602 - layer.2.k_cache 0.00125400 2.71079335 - layer.2.v_cache 0.00000109 0.00415155 - layer.3.k_cache 0.00146534 3.16813822 - layer.3.v_cache 0.00000206 0.00713544 - layer.4.k_cache 0.00325167 7.50724630 - layer.4.v_cache 0.00000289 0.01075022 - layer.4.output 0.00022000 0.31665731 - ------------------------------------------------------------------------------------- - TOTAL 0.00290784 16.52973827 - (elements=673,792) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 673792 -Total Bytes 10052 -BPFP 0.1193 bits/point -EBPFP 0.2387 equivalent bits/point -MSE 16.529738 ----------------------- -------------------------------------------------------- -Time: 0.356s Load: 0.004s, Pack+Encode: 0.155s, Decode+Unpack: 0.197s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 47, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 47, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 16.5297 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample1295-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample1295-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample13-layer4-item1.zst (9/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample13-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 154, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 154, 128) -Output shape: (1, 154, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.0.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.1.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.1.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.2.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.2.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.3.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.3.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.4.k_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.4.v_cache: torch.Size([1, 8, 154, 128]) -> torch.Size([1, 1, 154, 1024]) - layer.4.output: torch.Size([1, 154, 4096]) -> torch.Size([1, 1, 154, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,504B, BPFP=0.0763 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,428B, BPFP=0.0724 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,976B, BPFP=0.1002 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,648B, BPFP=0.0836 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,160B, BPFP=0.1096 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,796B, BPFP=0.0911 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,248B, BPFP=0.1140 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,924B, BPFP=0.0976 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,908B, BPFP=0.0968 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,892B, BPFP=0.0960 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 6,144B, BPFP=0.0779 -⌛️ [2/4] FRONTEND: Frontend time: 0.259s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 154, 128]) - layer.0.v_cache: torch.Size([1, 8, 154, 128]) - layer.1.k_cache: torch.Size([1, 8, 154, 128]) - layer.1.v_cache: torch.Size([1, 8, 154, 128]) - layer.2.k_cache: torch.Size([1, 8, 154, 128]) - layer.2.v_cache: torch.Size([1, 8, 154, 128]) - layer.3.k_cache: torch.Size([1, 8, 154, 128]) - layer.3.v_cache: torch.Size([1, 8, 154, 128]) - layer.4.k_cache: torch.Size([1, 8, 154, 128]) - layer.4.v_cache: torch.Size([1, 8, 154, 128]) - layer.4.output: torch.Size([1, 154, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.377s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 154, 128]) - layer.0.v_cache: torch.Size([1, 8, 154, 128]) - layer.1.k_cache: torch.Size([1, 8, 154, 128]) - layer.1.v_cache: torch.Size([1, 8, 154, 128]) - layer.2.k_cache: torch.Size([1, 8, 154, 128]) - layer.2.v_cache: torch.Size([1, 8, 154, 128]) - layer.3.k_cache: torch.Size([1, 8, 154, 128]) - layer.3.v_cache: torch.Size([1, 8, 154, 128]) - layer.4.k_cache: torch.Size([1, 8, 154, 128]) - layer.4.v_cache: torch.Size([1, 8, 154, 128]) - layer.4.output: torch.Size([1, 154, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02515895 95.00974026 - layer.0.v_cache 0.00000027 0.00065902 - layer.1.k_cache 0.00328501 6.15706010 - layer.1.v_cache 0.00000091 0.00262582 - layer.2.k_cache 0.00115449 2.12493877 - layer.2.v_cache 0.00000127 0.00420791 - layer.3.k_cache 0.00140159 2.57953674 - layer.3.v_cache 0.00000237 0.00755798 - layer.4.k_cache 0.00336527 6.49507438 - layer.4.v_cache 0.00000311 0.01179328 - layer.4.output 0.00016660 0.23228043 - ------------------------------------------------------------------------------------- - TOTAL 0.00250283 8.09445114 - (elements=2,207,744) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2207744 -Total Bytes 24628 -BPFP 0.0892 bits/point -EBPFP 0.1785 equivalent bits/point -MSE 8.094451 ----------------------- -------------------------------------------------------- -Time: 0.643s Load: 0.008s, Pack+Encode: 0.259s, Decode+Unpack: 0.377s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 154, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 154, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.0945 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample13-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample13-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample130-layer4-item1.zst (10/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample130-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 100, 128) -Output shape: (1, 100, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 988B, BPFP=0.0772 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,144B, BPFP=0.0894 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,428B, BPFP=0.1116 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,312B, BPFP=0.1025 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,532B, BPFP=0.1197 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,368B, BPFP=0.1069 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,500B, BPFP=0.1172 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,456B, BPFP=0.1138 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,372B, BPFP=0.1072 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,428B, BPFP=0.1116 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,164B, BPFP=0.0813 -⌛️ [2/4] FRONTEND: Frontend time: 0.205s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.output: torch.Size([1, 100, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.output: torch.Size([1, 100, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02583573 98.90086914 - layer.0.v_cache 0.00000027 0.00067624 - layer.1.k_cache 0.00330333 6.86186646 - layer.1.v_cache 0.00000088 0.00272235 - layer.2.k_cache 0.00114776 2.22495834 - layer.2.v_cache 0.00000114 0.00414716 - layer.3.k_cache 0.00139474 2.73910156 - layer.3.v_cache 0.00000229 0.00756324 - layer.4.k_cache 0.00328125 6.55173645 - layer.4.v_cache 0.00000316 0.01201035 - layer.4.output 0.00020624 0.27495237 - ------------------------------------------------------------------------------------- - TOTAL 0.00255682 8.45753291 - (elements=1,433,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1433600 -Total Bytes 17692 -BPFP 0.0987 bits/point -EBPFP 0.1975 equivalent bits/point -MSE 8.457533 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.007s, Pack+Encode: 0.205s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.4575 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample130-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample130-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample144-layer4-item1.zst (11/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample144-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 95, 128) -Output shape: (1, 95, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 972B, BPFP=0.0799 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,164B, BPFP=0.0957 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,412B, BPFP=0.1161 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,284B, BPFP=0.1056 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,540B, BPFP=0.1266 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,368B, BPFP=0.1125 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,500B, BPFP=0.1234 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,416B, BPFP=0.1164 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,380B, BPFP=0.1135 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,392B, BPFP=0.1145 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,820B, BPFP=0.0785 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02533972 93.22819696 - layer.0.v_cache 0.00000026 0.00066656 - layer.1.k_cache 0.00345234 6.72206582 - layer.1.v_cache 0.00000076 0.00260556 - layer.2.k_cache 0.00126989 2.22042590 - layer.2.v_cache 0.00000106 0.00394459 - layer.3.k_cache 0.00140666 2.65545381 - layer.3.v_cache 0.00000216 0.00717765 - layer.4.k_cache 0.00351815 6.53643735 - layer.4.v_cache 0.00000309 0.01178080 - layer.4.output 0.00017514 0.24736834 - ------------------------------------------------------------------------------------- - TOTAL 0.00254962 8.02701631 - (elements=1,361,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1361920 -Total Bytes 17248 -BPFP 0.1013 bits/point -EBPFP 0.2026 equivalent bits/point -MSE 8.027016 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.006s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.0270 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample144-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample144-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample145-layer4-item1.zst (12/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample145-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 952B, BPFP=0.0767 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,080B, BPFP=0.0870 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,388B, BPFP=0.1118 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,216B, BPFP=0.0979 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,508B, BPFP=0.1215 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,324B, BPFP=0.1066 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,564B, BPFP=0.1260 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,332B, BPFP=0.1073 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,324B, BPFP=0.1066 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,340B, BPFP=0.1079 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,628B, BPFP=0.0731 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02570412 127.82747060 - layer.0.v_cache 0.00000027 0.00066172 - layer.1.k_cache 0.00344234 6.73930689 - layer.1.v_cache 0.00000085 0.00257758 - layer.2.k_cache 0.00116154 2.23637846 - layer.2.v_cache 0.00000108 0.00393157 - layer.3.k_cache 0.00150460 2.73132639 - layer.3.v_cache 0.00000231 0.00690119 - layer.4.k_cache 0.00334272 6.67359028 - layer.4.v_cache 0.00000310 0.01133342 - layer.4.output 0.00020164 0.25707925 - ------------------------------------------------------------------------------------- - TOTAL 0.00256925 10.51869965 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 16656 -BPFP 0.0958 bits/point -EBPFP 0.1916 equivalent bits/point -MSE 10.518700 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.006s, Pack+Encode: 0.204s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.5187 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample145-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample145-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample146-layer4-item1.zst (13/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample146-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 98, 128) -Output shape: (1, 98, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 936B, BPFP=0.0746 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,016B, BPFP=0.0810 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,376B, BPFP=0.1097 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,308B, BPFP=0.1043 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,496B, BPFP=0.1193 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,356B, BPFP=0.1081 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,508B, BPFP=0.1202 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,464B, BPFP=0.1167 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,356B, BPFP=0.1081 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,392B, BPFP=0.1110 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,860B, BPFP=0.0769 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.output: torch.Size([1, 98, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.output: torch.Size([1, 98, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02660144 110.27688536 - layer.0.v_cache 0.00000028 0.00068844 - layer.1.k_cache 0.00330389 6.76735391 - layer.1.v_cache 0.00000100 0.00274468 - layer.2.k_cache 0.00113731 2.20570560 - layer.2.v_cache 0.00000122 0.00442053 - layer.3.k_cache 0.00133429 2.75479687 - layer.3.v_cache 0.00000236 0.00765428 - layer.4.k_cache 0.00329518 6.55432752 - layer.4.v_cache 0.00000318 0.01209114 - layer.4.output 0.00020743 0.27170934 - ------------------------------------------------------------------------------------- - TOTAL 0.00260785 9.26239327 - (elements=1,404,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1404928 -Total Bytes 17068 -BPFP 0.0972 bits/point -EBPFP 0.1944 equivalent bits/point -MSE 9.262393 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.007s, Pack+Encode: 0.203s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.2624 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample146-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample146-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample147-layer4-item1.zst (14/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample147-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 98, 128) -Output shape: (1, 98, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) -> torch.Size([1, 1, 98, 1024]) - layer.4.output: torch.Size([1, 98, 4096]) -> torch.Size([1, 1, 98, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 940B, BPFP=0.0749 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,076B, BPFP=0.0858 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,392B, BPFP=0.1110 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,224B, BPFP=0.0976 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,528B, BPFP=0.1218 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,356B, BPFP=0.1081 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,540B, BPFP=0.1228 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,400B, BPFP=0.1116 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,364B, BPFP=0.1087 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,396B, BPFP=0.1113 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,844B, BPFP=0.0766 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.output: torch.Size([1, 98, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 98, 128]) - layer.0.v_cache: torch.Size([1, 8, 98, 128]) - layer.1.k_cache: torch.Size([1, 8, 98, 128]) - layer.1.v_cache: torch.Size([1, 8, 98, 128]) - layer.2.k_cache: torch.Size([1, 8, 98, 128]) - layer.2.v_cache: torch.Size([1, 8, 98, 128]) - layer.3.k_cache: torch.Size([1, 8, 98, 128]) - layer.3.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.k_cache: torch.Size([1, 8, 98, 128]) - layer.4.v_cache: torch.Size([1, 8, 98, 128]) - layer.4.output: torch.Size([1, 98, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02655566 107.38740633 - layer.0.v_cache 0.00000028 0.00069665 - layer.1.k_cache 0.00343622 6.97199483 - layer.1.v_cache 0.00000088 0.00272382 - layer.2.k_cache 0.00122497 2.25929416 - layer.2.v_cache 0.00000114 0.00425762 - layer.3.k_cache 0.00135168 2.70744573 - layer.3.v_cache 0.00000224 0.00748202 - layer.4.k_cache 0.00342672 6.75755279 - layer.4.v_cache 0.00000300 0.01150348 - layer.4.output 0.00022158 0.25217457 - ------------------------------------------------------------------------------------- - TOTAL 0.00263494 9.07993255 - (elements=1,404,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1404928 -Total Bytes 17060 -BPFP 0.0971 bits/point -EBPFP 0.1943 equivalent bits/point -MSE 9.079933 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 98, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 98, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.0799 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample147-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample147-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample150-layer4-item1.zst (15/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample150-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 100, 128) -Output shape: (1, 100, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) -> torch.Size([1, 1, 100, 1024]) - layer.4.output: torch.Size([1, 100, 4096]) -> torch.Size([1, 1, 100, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 996B, BPFP=0.0778 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,092B, BPFP=0.0853 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,424B, BPFP=0.1113 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,252B, BPFP=0.0978 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,508B, BPFP=0.1178 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,348B, BPFP=0.1053 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,504B, BPFP=0.1175 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,436B, BPFP=0.1122 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,332B, BPFP=0.1041 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,436B, BPFP=0.1122 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,920B, BPFP=0.0766 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.output: torch.Size([1, 100, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 100, 128]) - layer.0.v_cache: torch.Size([1, 8, 100, 128]) - layer.1.k_cache: torch.Size([1, 8, 100, 128]) - layer.1.v_cache: torch.Size([1, 8, 100, 128]) - layer.2.k_cache: torch.Size([1, 8, 100, 128]) - layer.2.v_cache: torch.Size([1, 8, 100, 128]) - layer.3.k_cache: torch.Size([1, 8, 100, 128]) - layer.3.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.k_cache: torch.Size([1, 8, 100, 128]) - layer.4.v_cache: torch.Size([1, 8, 100, 128]) - layer.4.output: torch.Size([1, 100, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02664840 100.31947266 - layer.0.v_cache 0.00000027 0.00067590 - layer.1.k_cache 0.00335783 6.71226929 - layer.1.v_cache 0.00000086 0.00269999 - layer.2.k_cache 0.00109031 2.21022644 - layer.2.v_cache 0.00000115 0.00417532 - layer.3.k_cache 0.00136642 2.68096802 - layer.3.v_cache 0.00000222 0.00728539 - layer.4.k_cache 0.00341075 6.80462708 - layer.4.v_cache 0.00000317 0.01155727 - layer.4.output 0.00031254 0.26127817 - ------------------------------------------------------------------------------------- - TOTAL 0.00265226 8.55707643 - (elements=1,433,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1433600 -Total Bytes 17248 -BPFP 0.0963 bits/point -EBPFP 0.1925 equivalent bits/point -MSE 8.557076 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 100, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 100, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.5571 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample150-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample150-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample153-layer4-item1.zst (16/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample153-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 964B, BPFP=0.0776 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,072B, BPFP=0.0863 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,384B, BPFP=0.1115 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,260B, BPFP=0.1015 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,544B, BPFP=0.1244 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,312B, BPFP=0.1057 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,528B, BPFP=0.1231 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,368B, BPFP=0.1102 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,368B, BPFP=0.1102 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,404B, BPFP=0.1131 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,632B, BPFP=0.0731 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02654818 133.09003504 - layer.0.v_cache 0.00000027 0.00067386 - layer.1.k_cache 0.00347710 6.80716037 - layer.1.v_cache 0.00000089 0.00283655 - layer.2.k_cache 0.00114081 2.24750605 - layer.2.v_cache 0.00000111 0.00417822 - layer.3.k_cache 0.00140715 2.73640646 - layer.3.v_cache 0.00000211 0.00718578 - layer.4.k_cache 0.00324570 6.50914835 - layer.4.v_cache 0.00000329 0.01202858 - layer.4.output 0.00023085 0.26058311 - ------------------------------------------------------------------------------------- - TOTAL 0.00262500 10.88996369 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 16836 -BPFP 0.0969 bits/point -EBPFP 0.1937 equivalent bits/point -MSE 10.889964 ----------------------- -------------------------------------------------------- -Time: 0.494s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.8900 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample153-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample153-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample154-layer4-item1.zst (17/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample154-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 89, 128) -Output shape: (1, 89, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.output: torch.Size([1, 89, 4096]) -> torch.Size([1, 1, 89, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 964B, BPFP=0.0846 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,028B, BPFP=0.0902 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,400B, BPFP=0.1229 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,288B, BPFP=0.1131 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,480B, BPFP=0.1299 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,296B, BPFP=0.1138 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,548B, BPFP=0.1359 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,408B, BPFP=0.1236 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,344B, BPFP=0.1180 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,376B, BPFP=0.1208 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,760B, BPFP=0.0825 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02697376 137.66814431 - layer.0.v_cache 0.00000027 0.00065884 - layer.1.k_cache 0.00348241 6.57572337 - layer.1.v_cache 0.00000086 0.00258880 - layer.2.k_cache 0.00112258 2.15763666 - layer.2.v_cache 0.00000113 0.00403265 - layer.3.k_cache 0.00136851 2.68311627 - layer.3.v_cache 0.00000223 0.00696568 - layer.4.k_cache 0.00328884 6.72641848 - layer.4.v_cache 0.00000300 0.01117885 - layer.4.output 0.00019035 0.26261874 - ------------------------------------------------------------------------------------- - TOTAL 0.00264321 11.20620992 - (elements=1,275,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1275904 -Total Bytes 16892 -BPFP 0.1059 bits/point -EBPFP 0.2118 equivalent bits/point -MSE 11.206210 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.006s, Pack+Encode: 0.204s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.2062 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample154-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample154-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample157-layer4-item1.zst (18/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample157-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 104, 128) -Output shape: (1, 104, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.0.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.1.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.1.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.2.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.2.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.3.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.3.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.4.k_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.4.v_cache: torch.Size([1, 8, 104, 128]) -> torch.Size([1, 1, 104, 1024]) - layer.4.output: torch.Size([1, 104, 4096]) -> torch.Size([1, 1, 104, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,100B, BPFP=0.0826 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,192B, BPFP=0.0895 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,516B, BPFP=0.1139 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,336B, BPFP=0.1004 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,552B, BPFP=0.1166 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,436B, BPFP=0.1079 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,588B, BPFP=0.1193 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,484B, BPFP=0.1115 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,404B, BPFP=0.1055 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,456B, BPFP=0.1094 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,196B, BPFP=0.0788 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 104, 128]) - layer.0.v_cache: torch.Size([1, 8, 104, 128]) - layer.1.k_cache: torch.Size([1, 8, 104, 128]) - layer.1.v_cache: torch.Size([1, 8, 104, 128]) - layer.2.k_cache: torch.Size([1, 8, 104, 128]) - layer.2.v_cache: torch.Size([1, 8, 104, 128]) - layer.3.k_cache: torch.Size([1, 8, 104, 128]) - layer.3.v_cache: torch.Size([1, 8, 104, 128]) - layer.4.k_cache: torch.Size([1, 8, 104, 128]) - layer.4.v_cache: torch.Size([1, 8, 104, 128]) - layer.4.output: torch.Size([1, 104, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 104, 128]) - layer.0.v_cache: torch.Size([1, 8, 104, 128]) - layer.1.k_cache: torch.Size([1, 8, 104, 128]) - layer.1.v_cache: torch.Size([1, 8, 104, 128]) - layer.2.k_cache: torch.Size([1, 8, 104, 128]) - layer.2.v_cache: torch.Size([1, 8, 104, 128]) - layer.3.k_cache: torch.Size([1, 8, 104, 128]) - layer.3.v_cache: torch.Size([1, 8, 104, 128]) - layer.4.k_cache: torch.Size([1, 8, 104, 128]) - layer.4.v_cache: torch.Size([1, 8, 104, 128]) - layer.4.output: torch.Size([1, 104, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02512770 74.87982647 - layer.0.v_cache 0.00000027 0.00068278 - layer.1.k_cache 0.00323274 6.54020632 - layer.1.v_cache 0.00000093 0.00283878 - layer.2.k_cache 0.00116098 2.18258946 - layer.2.v_cache 0.00000116 0.00427826 - layer.3.k_cache 0.00137668 2.67482317 - layer.3.v_cache 0.00000213 0.00726331 - layer.4.k_cache 0.00339989 6.71496289 - layer.4.v_cache 0.00000305 0.01162549 - layer.4.output 0.00020049 0.22826228 - ------------------------------------------------------------------------------------- - TOTAL 0.00250768 6.70943900 - (elements=1,490,944) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1490944 -Total Bytes 18260 -BPFP 0.0980 bits/point -EBPFP 0.1960 equivalent bits/point -MSE 6.709439 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 104, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 104, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.7094 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample157-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample157-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample159-layer4-item1.zst (19/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample159-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 82, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 82, 128) -Output shape: (1, 82, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.output: torch.Size([1, 82, 4096]) -> torch.Size([1, 1, 82, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 920B, BPFP=0.0877 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 940B, BPFP=0.0896 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,300B, BPFP=0.1239 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,084B, BPFP=0.1033 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,368B, BPFP=0.1303 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,172B, BPFP=0.1117 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1372 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,248B, BPFP=0.1189 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,192B, BPFP=0.1136 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,172B, BPFP=0.1117 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,508B, BPFP=0.0836 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.output: torch.Size([1, 82, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.output: torch.Size([1, 82, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02805163 91.97551448 - layer.0.v_cache 0.00000027 0.00063564 - layer.1.k_cache 0.00355746 6.71872246 - layer.1.v_cache 0.00000072 0.00239335 - layer.2.k_cache 0.00112546 2.18650967 - layer.2.v_cache 0.00000104 0.00377914 - layer.3.k_cache 0.00143209 2.68842204 - layer.3.v_cache 0.00000205 0.00655526 - layer.4.k_cache 0.00323843 6.22821715 - layer.4.v_cache 0.00000284 0.01050624 - layer.4.output 0.00022045 0.27187815 - ------------------------------------------------------------------------------------- - TOTAL 0.00273527 7.92205486 - (elements=1,175,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1175552 -Total Bytes 15344 -BPFP 0.1044 bits/point -EBPFP 0.2088 equivalent bits/point -MSE 7.922055 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.006s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 82, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.9221 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample159-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample159-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample165-layer4-item1.zst (20/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample165-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 109, 128) -Output shape: (1, 109, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) -> torch.Size([1, 1, 109, 1024]) - layer.4.output: torch.Size([1, 109, 4096]) -> torch.Size([1, 1, 109, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,080B, BPFP=0.0774 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,236B, BPFP=0.0886 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,528B, BPFP=0.1095 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,352B, BPFP=0.0969 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,552B, BPFP=0.1112 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,504B, BPFP=0.1078 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,612B, BPFP=0.1155 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,528B, BPFP=0.1095 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,436B, BPFP=0.1029 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,472B, BPFP=0.1055 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,040B, BPFP=0.0724 -⌛️ [2/4] FRONTEND: Frontend time: 0.214s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.output: torch.Size([1, 109, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.285s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 109, 128]) - layer.0.v_cache: torch.Size([1, 8, 109, 128]) - layer.1.k_cache: torch.Size([1, 8, 109, 128]) - layer.1.v_cache: torch.Size([1, 8, 109, 128]) - layer.2.k_cache: torch.Size([1, 8, 109, 128]) - layer.2.v_cache: torch.Size([1, 8, 109, 128]) - layer.3.k_cache: torch.Size([1, 8, 109, 128]) - layer.3.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.k_cache: torch.Size([1, 8, 109, 128]) - layer.4.v_cache: torch.Size([1, 8, 109, 128]) - layer.4.output: torch.Size([1, 109, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02442908 101.24989249 - layer.0.v_cache 0.00000028 0.00069334 - layer.1.k_cache 0.00329736 6.77860645 - layer.1.v_cache 0.00000085 0.00267007 - layer.2.k_cache 0.00114487 2.25178164 - layer.2.v_cache 0.00000123 0.00421611 - layer.3.k_cache 0.00134342 2.76778811 - layer.3.v_cache 0.00000206 0.00715152 - layer.4.k_cache 0.00348010 7.02449358 - layer.4.v_cache 0.00000308 0.01154098 - layer.4.output 0.00017639 0.24502256 - ------------------------------------------------------------------------------------- - TOTAL 0.00245771 8.64849461 - (elements=1,562,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1562624 -Total Bytes 18340 -BPFP 0.0939 bits/point -EBPFP 0.1878 equivalent bits/point -MSE 8.648495 ----------------------- -------------------------------------------------------- -Time: 0.507s Load: 0.008s, Pack+Encode: 0.214s, Decode+Unpack: 0.285s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 109, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 109, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.6485 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample165-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample165-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample167-layer4-item1.zst (21/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample167-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 95, 128) -Output shape: (1, 95, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 964B, BPFP=0.0793 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,088B, BPFP=0.0895 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,376B, BPFP=0.1132 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,200B, BPFP=0.0987 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,460B, BPFP=0.1201 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,316B, BPFP=0.1082 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,616B, BPFP=0.1329 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,444B, BPFP=0.1187 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,364B, BPFP=0.1122 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,428B, BPFP=0.1174 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,996B, BPFP=0.0822 -⌛️ [2/4] FRONTEND: Frontend time: 0.214s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02733303 92.13532072 - layer.0.v_cache 0.00000028 0.00068010 - layer.1.k_cache 0.00367247 6.70570582 - layer.1.v_cache 0.00000082 0.00267160 - layer.2.k_cache 0.00112740 2.16361662 - layer.2.v_cache 0.00000120 0.00425675 - layer.3.k_cache 0.00137809 2.68220568 - layer.3.v_cache 0.00000227 0.00735702 - layer.4.k_cache 0.00319271 6.40259303 - layer.4.v_cache 0.00000332 0.01212407 - layer.4.output 0.00022674 0.24830858 - ------------------------------------------------------------------------------------- - TOTAL 0.00268704 7.93641184 - (elements=1,361,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1361920 -Total Bytes 17252 -BPFP 0.1013 bits/point -EBPFP 0.2027 equivalent bits/point -MSE 7.936412 ----------------------- -------------------------------------------------------- -Time: 0.499s Load: 0.006s, Pack+Encode: 0.214s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.9364 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample167-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample167-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample168-layer4-item1.zst (22/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample168-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 103, 128) -Output shape: (1, 103, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,072B, BPFP=0.0813 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,232B, BPFP=0.0934 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,500B, BPFP=0.1138 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,336B, BPFP=0.1013 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,532B, BPFP=0.1162 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,424B, BPFP=0.1080 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,580B, BPFP=0.1198 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,484B, BPFP=0.1126 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,396B, BPFP=0.1059 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,436B, BPFP=0.1089 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,100B, BPFP=0.0777 -⌛️ [2/4] FRONTEND: Frontend time: 0.209s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.274s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.03002706 93.22386984 - layer.0.v_cache 0.00000028 0.00067966 - layer.1.k_cache 0.00329593 6.62200631 - layer.1.v_cache 0.00000087 0.00273607 - layer.2.k_cache 0.00114743 2.22888154 - layer.2.v_cache 0.00000126 0.00423294 - layer.3.k_cache 0.00133553 2.69756606 - layer.3.v_cache 0.00000217 0.00729059 - layer.4.k_cache 0.00338515 6.72401028 - layer.4.v_cache 0.00000318 0.01163243 - layer.4.output 0.00017408 0.22899083 - ------------------------------------------------------------------------------------- - TOTAL 0.00284966 8.03134779 - (elements=1,476,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1476608 -Total Bytes 18092 -BPFP 0.0980 bits/point -EBPFP 0.1960 equivalent bits/point -MSE 8.031348 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.008s, Pack+Encode: 0.209s, Decode+Unpack: 0.274s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.0313 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample168-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample168-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample174-layer4-item1.zst (23/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample174-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 103, 128) -Output shape: (1, 103, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,076B, BPFP=0.0816 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,164B, BPFP=0.0883 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,476B, BPFP=0.1120 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,332B, BPFP=0.1010 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,536B, BPFP=0.1165 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,432B, BPFP=0.1086 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,564B, BPFP=0.1186 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,504B, BPFP=0.1141 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,416B, BPFP=0.1074 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,432B, BPFP=0.1086 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,232B, BPFP=0.0802 -⌛️ [2/4] FRONTEND: Frontend time: 0.198s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.272s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02636531 97.47188827 - layer.0.v_cache 0.00000028 0.00068117 - layer.1.k_cache 0.00340973 6.68699216 - layer.1.v_cache 0.00000088 0.00265875 - layer.2.k_cache 0.00113950 2.18829494 - layer.2.v_cache 0.00000120 0.00421328 - layer.3.k_cache 0.00133761 2.69576671 - layer.3.v_cache 0.00000227 0.00744471 - layer.4.k_cache 0.00339299 6.73344466 - layer.4.v_cache 0.00000316 0.01188066 - layer.4.output 0.00023648 0.24492114 - ------------------------------------------------------------------------------------- - TOTAL 0.00261420 8.34163928 - (elements=1,476,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1476608 -Total Bytes 18164 -BPFP 0.0984 bits/point -EBPFP 0.1968 equivalent bits/point -MSE 8.341639 ----------------------- -------------------------------------------------------- -Time: 0.478s Load: 0.007s, Pack+Encode: 0.198s, Decode+Unpack: 0.272s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.3416 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample174-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample174-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample182-layer4-item1.zst (24/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample182-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 92, 128) -Output shape: (1, 92, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.output: torch.Size([1, 92, 4096]) -> torch.Size([1, 1, 92, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,012B, BPFP=0.0859 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,076B, BPFP=0.0914 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,412B, BPFP=0.1199 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,300B, BPFP=0.1104 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,528B, BPFP=0.1298 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,392B, BPFP=0.1182 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,576B, BPFP=0.1338 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,500B, BPFP=0.1274 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,400B, BPFP=0.1189 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,396B, BPFP=0.1185 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,020B, BPFP=0.0853 -⌛️ [2/4] FRONTEND: Frontend time: 0.197s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.272s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02458628 86.08865489 - layer.0.v_cache 0.00000028 0.00066088 - layer.1.k_cache 0.00347829 6.82534127 - layer.1.v_cache 0.00000086 0.00270307 - layer.2.k_cache 0.00113727 2.13102722 - layer.2.v_cache 0.00000116 0.00406760 - layer.3.k_cache 0.00132770 2.65583901 - layer.3.v_cache 0.00000243 0.00728937 - layer.4.k_cache 0.00339865 6.45119841 - layer.4.v_cache 0.00000297 0.01102581 - layer.4.output 0.00018871 0.24500965 - ------------------------------------------------------------------------------------- - TOTAL 0.00247791 7.51127472 - (elements=1,318,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1318912 -Total Bytes 17612 -BPFP 0.1068 bits/point -EBPFP 0.2137 equivalent bits/point -MSE 7.511275 ----------------------- -------------------------------------------------------- -Time: 0.475s Load: 0.006s, Pack+Encode: 0.197s, Decode+Unpack: 0.272s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.5113 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample182-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample182-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample185-layer4-item1.zst (25/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample185-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 96, 128) -Output shape: (1, 96, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.output: torch.Size([1, 96, 4096]) -> torch.Size([1, 1, 96, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 976B, BPFP=0.0794 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,000B, BPFP=0.0814 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,392B, BPFP=0.1133 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,176B, BPFP=0.0957 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,504B, BPFP=0.1224 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,260B, BPFP=0.1025 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,588B, BPFP=0.1292 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,340B, BPFP=0.1090 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,380B, BPFP=0.1123 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,364B, BPFP=0.1110 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,836B, BPFP=0.0780 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02634023 147.04760742 - layer.0.v_cache 0.00000027 0.00066454 - layer.1.k_cache 0.00351230 6.96323013 - layer.1.v_cache 0.00000075 0.00256841 - layer.2.k_cache 0.00114531 2.25685469 - layer.2.v_cache 0.00000110 0.00400287 - layer.3.k_cache 0.00136477 2.70772235 - layer.3.v_cache 0.00000217 0.00732755 - layer.4.k_cache 0.00356451 6.99774170 - layer.4.v_cache 0.00000291 0.01120134 - layer.4.output 0.00018809 0.26704268 - ------------------------------------------------------------------------------------- - TOTAL 0.00262048 11.93336369 - (elements=1,376,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1376256 -Total Bytes 16816 -BPFP 0.0977 bits/point -EBPFP 0.1955 equivalent bits/point -MSE 11.933364 ----------------------- -------------------------------------------------------- -Time: 0.488s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.9334 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample185-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample185-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample191-layer4-item1.zst (26/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample191-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 93, 128) -Output shape: (1, 93, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.output: torch.Size([1, 93, 4096]) -> torch.Size([1, 1, 93, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,000B, BPFP=0.0840 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,116B, BPFP=0.0938 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,400B, BPFP=0.1176 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,264B, BPFP=0.1062 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,560B, BPFP=0.1310 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,372B, BPFP=0.1153 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,560B, BPFP=0.1310 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,456B, BPFP=0.1223 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,408B, BPFP=0.1183 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,472B, BPFP=0.1237 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,084B, BPFP=0.0858 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02618607 81.37398143 - layer.0.v_cache 0.00000028 0.00066692 - layer.1.k_cache 0.00358459 6.52884928 - layer.1.v_cache 0.00000078 0.00264363 - layer.2.k_cache 0.00113483 2.15311653 - layer.2.v_cache 0.00000114 0.00412703 - layer.3.k_cache 0.00136506 2.66358899 - layer.3.v_cache 0.00000214 0.00715240 - layer.4.k_cache 0.00334622 6.28350239 - layer.4.v_cache 0.00000303 0.01159796 - layer.4.output 0.00022383 0.25347367 - ------------------------------------------------------------------------------------- - TOTAL 0.00260854 7.14593723 - (elements=1,333,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1333248 -Total Bytes 17692 -BPFP 0.1062 bits/point -EBPFP 0.2123 equivalent bits/point -MSE 7.145937 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.1459 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample191-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample191-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample196-layer4-item1.zst (27/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample196-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 96, 128) -Output shape: (1, 96, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.output: torch.Size([1, 96, 4096]) -> torch.Size([1, 1, 96, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 980B, BPFP=0.0798 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,132B, BPFP=0.0921 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,412B, BPFP=0.1149 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,272B, BPFP=0.1035 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,556B, BPFP=0.1266 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,344B, BPFP=0.1094 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,576B, BPFP=0.1283 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,516B, BPFP=0.1234 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,396B, BPFP=0.1136 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,448B, BPFP=0.1178 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,020B, BPFP=0.0818 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02641018 146.77142334 - layer.0.v_cache 0.00000028 0.00068233 - layer.1.k_cache 0.00334712 7.03527768 - layer.1.v_cache 0.00000079 0.00281030 - layer.2.k_cache 0.00118939 2.26469262 - layer.2.v_cache 0.00000116 0.00444049 - layer.3.k_cache 0.00138481 2.74837081 - layer.3.v_cache 0.00000224 0.00754625 - layer.4.k_cache 0.00336247 6.82258987 - layer.4.v_cache 0.00000319 0.01186468 - layer.4.output 0.00019453 0.23430779 - ------------------------------------------------------------------------------------- - TOTAL 0.00260570 11.90049497 - (elements=1,376,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1376256 -Total Bytes 17652 -BPFP 0.1026 bits/point -EBPFP 0.2052 equivalent bits/point -MSE 11.900495 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.9005 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample196-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample196-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample197-layer4-item1.zst (28/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample197-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 92, 128) -Output shape: (1, 92, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.output: torch.Size([1, 92, 4096]) -> torch.Size([1, 1, 92, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,004B, BPFP=0.0853 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,052B, BPFP=0.0893 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,376B, BPFP=0.1168 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,288B, BPFP=0.1094 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,528B, BPFP=0.1298 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,372B, BPFP=0.1165 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,532B, BPFP=0.1301 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,392B, BPFP=0.1182 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,376B, BPFP=0.1168 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,396B, BPFP=0.1185 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,884B, BPFP=0.0825 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02542763 85.26472274 - layer.0.v_cache 0.00000028 0.00065769 - layer.1.k_cache 0.00330708 6.61761276 - layer.1.v_cache 0.00000075 0.00268580 - layer.2.k_cache 0.00113337 2.19234865 - layer.2.v_cache 0.00000121 0.00429697 - layer.3.k_cache 0.00137180 2.65683448 - layer.3.v_cache 0.00000212 0.00717042 - layer.4.k_cache 0.00335567 6.53588469 - layer.4.v_cache 0.00000315 0.01169791 - layer.4.output 0.00020664 0.24448413 - ------------------------------------------------------------------------------------- - TOTAL 0.00253069 7.44798919 - (elements=1,318,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1318912 -Total Bytes 17200 -BPFP 0.1043 bits/point -EBPFP 0.2087 equivalent bits/point -MSE 7.447989 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.4480 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample197-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample197-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample2-layer4-item1.zst (29/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample2-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 178, 128) -Output shape: (1, 178, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) -> torch.Size([1, 1, 178, 1024]) - layer.4.output: torch.Size([1, 178, 4096]) -> torch.Size([1, 1, 178, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,632B, BPFP=0.0716 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,640B, BPFP=0.0720 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,236B, BPFP=0.0981 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,912B, BPFP=0.0839 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,276B, BPFP=0.0999 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,064B, BPFP=0.0906 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,368B, BPFP=0.1039 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,152B, BPFP=0.0945 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,116B, BPFP=0.0929 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,104B, BPFP=0.0923 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 6,060B, BPFP=0.0665 -⌛️ [2/4] FRONTEND: Frontend time: 0.253s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.377s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 178, 128]) - layer.0.v_cache: torch.Size([1, 8, 178, 128]) - layer.1.k_cache: torch.Size([1, 8, 178, 128]) - layer.1.v_cache: torch.Size([1, 8, 178, 128]) - layer.2.k_cache: torch.Size([1, 8, 178, 128]) - layer.2.v_cache: torch.Size([1, 8, 178, 128]) - layer.3.k_cache: torch.Size([1, 8, 178, 128]) - layer.3.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.k_cache: torch.Size([1, 8, 178, 128]) - layer.4.v_cache: torch.Size([1, 8, 178, 128]) - layer.4.output: torch.Size([1, 178, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02409235 62.94261324 - layer.0.v_cache 0.00000026 0.00064029 - layer.1.k_cache 0.00309449 6.00199633 - layer.1.v_cache 0.00000082 0.00252857 - layer.2.k_cache 0.00118451 2.04860713 - layer.2.v_cache 0.00000118 0.00397057 - layer.3.k_cache 0.00136933 2.52307866 - layer.3.v_cache 0.00000226 0.00721894 - layer.4.k_cache 0.00340187 6.31720931 - layer.4.v_cache 0.00000300 0.01112392 - layer.4.output 0.00017401 0.21673335 - ------------------------------------------------------------------------------------- - TOTAL 0.00241758 5.76613717 - (elements=2,551,808) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2551808 -Total Bytes 26560 -BPFP 0.0833 bits/point -EBPFP 0.1665 equivalent bits/point -MSE 5.766137 ----------------------- -------------------------------------------------------- -Time: 0.640s Load: 0.009s, Pack+Encode: 0.253s, Decode+Unpack: 0.377s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 178, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 178, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.7661 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample2-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample2-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample201-layer4-item1.zst (30/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample201-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 95, 128) -Output shape: (1, 95, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) -> torch.Size([1, 1, 95, 1024]) - layer.4.output: torch.Size([1, 95, 4096]) -> torch.Size([1, 1, 95, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 960B, BPFP=0.0789 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,032B, BPFP=0.0849 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,376B, BPFP=0.1132 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,208B, BPFP=0.0993 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,496B, BPFP=0.1230 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,388B, BPFP=0.1141 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,516B, BPFP=0.1247 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,392B, BPFP=0.1145 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,364B, BPFP=0.1122 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,388B, BPFP=0.1141 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,800B, BPFP=0.0781 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 95, 128]) - layer.0.v_cache: torch.Size([1, 8, 95, 128]) - layer.1.k_cache: torch.Size([1, 8, 95, 128]) - layer.1.v_cache: torch.Size([1, 8, 95, 128]) - layer.2.k_cache: torch.Size([1, 8, 95, 128]) - layer.2.v_cache: torch.Size([1, 8, 95, 128]) - layer.3.k_cache: torch.Size([1, 8, 95, 128]) - layer.3.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.k_cache: torch.Size([1, 8, 95, 128]) - layer.4.v_cache: torch.Size([1, 8, 95, 128]) - layer.4.output: torch.Size([1, 95, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02560628 96.74490132 - layer.0.v_cache 0.00000028 0.00067098 - layer.1.k_cache 0.00332260 6.68604993 - layer.1.v_cache 0.00000077 0.00258034 - layer.2.k_cache 0.00114542 2.21772702 - layer.2.v_cache 0.00000120 0.00408465 - layer.3.k_cache 0.00143781 2.70086349 - layer.3.v_cache 0.00000216 0.00684829 - layer.4.k_cache 0.00324414 6.44055883 - layer.4.v_cache 0.00000298 0.01106810 - layer.4.output 0.00027467 0.27511665 - ------------------------------------------------------------------------------------- - TOTAL 0.00256159 8.27970140 - (elements=1,361,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1361920 -Total Bytes 16920 -BPFP 0.0994 bits/point -EBPFP 0.1988 equivalent bits/point -MSE 8.279701 ----------------------- -------------------------------------------------------- -Time: 0.494s Load: 0.006s, Pack+Encode: 0.204s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 95, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 95, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.2797 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample201-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample201-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample212-layer4-item1.zst (31/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample212-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 94, 128) -Output shape: (1, 94, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.output: torch.Size([1, 94, 4096]) -> torch.Size([1, 1, 94, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 980B, BPFP=0.0814 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,044B, BPFP=0.0868 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,372B, BPFP=0.1140 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,188B, BPFP=0.0987 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,492B, BPFP=0.1240 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,272B, BPFP=0.1057 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,524B, BPFP=0.1267 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,420B, BPFP=0.1180 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,384B, BPFP=0.1150 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,388B, BPFP=0.1154 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,812B, BPFP=0.0792 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02535593 76.80986640 - layer.0.v_cache 0.00000029 0.00067027 - layer.1.k_cache 0.00347189 6.61654144 - layer.1.v_cache 0.00000073 0.00249481 - layer.2.k_cache 0.00115370 2.23407080 - layer.2.v_cache 0.00000109 0.00397873 - layer.3.k_cache 0.00136415 2.64650337 - layer.3.v_cache 0.00000218 0.00701761 - layer.4.k_cache 0.00350239 6.72675599 - layer.4.v_cache 0.00000301 0.01125667 - layer.4.output 0.00017689 0.24229378 - ------------------------------------------------------------------------------------- - TOTAL 0.00254021 6.85916652 - (elements=1,347,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1347584 -Total Bytes 16876 -BPFP 0.1002 bits/point -EBPFP 0.2004 equivalent bits/point -MSE 6.859167 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.8592 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample212-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample212-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample214-layer4-item1.zst (32/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample214-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 99, 128) -Output shape: (1, 99, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 964B, BPFP=0.0761 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,156B, BPFP=0.0912 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,384B, BPFP=0.1092 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,228B, BPFP=0.0969 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,508B, BPFP=0.1190 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,364B, BPFP=0.1076 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,548B, BPFP=0.1222 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,428B, BPFP=0.1127 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,356B, BPFP=0.1070 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,376B, BPFP=0.1086 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,860B, BPFP=0.0762 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.output: torch.Size([1, 99, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.output: torch.Size([1, 99, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02440176 121.90934738 - layer.0.v_cache 0.00000028 0.00066899 - layer.1.k_cache 0.00326434 6.91836147 - layer.1.v_cache 0.00000081 0.00259468 - layer.2.k_cache 0.00115982 2.28968580 - layer.2.v_cache 0.00000131 0.00421555 - layer.3.k_cache 0.00139121 2.75908715 - layer.3.v_cache 0.00000205 0.00706094 - layer.4.k_cache 0.00335893 6.79332140 - layer.4.v_cache 0.00000299 0.01142833 - layer.4.output 0.00020438 0.25125203 - ------------------------------------------------------------------------------------- - TOTAL 0.00245722 10.12148427 - (elements=1,419,264) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1419264 -Total Bytes 17172 -BPFP 0.0968 bits/point -EBPFP 0.1936 equivalent bits/point -MSE 10.121484 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.006s, Pack+Encode: 0.204s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.1215 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample214-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample214-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample224-layer4-item1.zst (33/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample224-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 103, 128) -Output shape: (1, 103, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) -> torch.Size([1, 1, 103, 1024]) - layer.4.output: torch.Size([1, 103, 4096]) -> torch.Size([1, 1, 103, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,064B, BPFP=0.0807 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,148B, BPFP=0.0871 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,464B, BPFP=0.1110 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,272B, BPFP=0.0965 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,516B, BPFP=0.1150 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,408B, BPFP=0.1068 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,568B, BPFP=0.1189 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,464B, BPFP=0.1110 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,360B, BPFP=0.1032 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,484B, BPFP=0.1126 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,160B, BPFP=0.0789 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 103, 128]) - layer.0.v_cache: torch.Size([1, 8, 103, 128]) - layer.1.k_cache: torch.Size([1, 8, 103, 128]) - layer.1.v_cache: torch.Size([1, 8, 103, 128]) - layer.2.k_cache: torch.Size([1, 8, 103, 128]) - layer.2.v_cache: torch.Size([1, 8, 103, 128]) - layer.3.k_cache: torch.Size([1, 8, 103, 128]) - layer.3.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.k_cache: torch.Size([1, 8, 103, 128]) - layer.4.v_cache: torch.Size([1, 8, 103, 128]) - layer.4.output: torch.Size([1, 103, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02603974 98.35331652 - layer.0.v_cache 0.00000026 0.00066323 - layer.1.k_cache 0.00325535 6.64213488 - layer.1.v_cache 0.00000093 0.00259113 - layer.2.k_cache 0.00112553 2.14156379 - layer.2.v_cache 0.00000114 0.00393924 - layer.3.k_cache 0.00141218 2.64621505 - layer.3.v_cache 0.00000204 0.00692432 - layer.4.k_cache 0.00324713 6.64003243 - layer.4.v_cache 0.00000302 0.01114293 - layer.4.output 0.00020093 0.26829529 - ------------------------------------------------------------------------------------- - TOTAL 0.00256365 8.39440748 - (elements=1,476,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1476608 -Total Bytes 17908 -BPFP 0.0970 bits/point -EBPFP 0.1940 equivalent bits/point -MSE 8.394407 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 103, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 103, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.3944 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample224-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample224-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample227-layer4-item1.zst (34/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample227-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 96, 128) -Output shape: (1, 96, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) -> torch.Size([1, 1, 96, 1024]) - layer.4.output: torch.Size([1, 96, 4096]) -> torch.Size([1, 1, 96, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 980B, BPFP=0.0798 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,048B, BPFP=0.0853 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,388B, BPFP=0.1130 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,240B, BPFP=0.1009 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,472B, BPFP=0.1198 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,336B, BPFP=0.1087 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,580B, BPFP=0.1286 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,384B, BPFP=0.1126 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,384B, BPFP=0.1126 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,356B, BPFP=0.1104 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,900B, BPFP=0.0793 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 96, 128]) - layer.0.v_cache: torch.Size([1, 8, 96, 128]) - layer.1.k_cache: torch.Size([1, 8, 96, 128]) - layer.1.v_cache: torch.Size([1, 8, 96, 128]) - layer.2.k_cache: torch.Size([1, 8, 96, 128]) - layer.2.v_cache: torch.Size([1, 8, 96, 128]) - layer.3.k_cache: torch.Size([1, 8, 96, 128]) - layer.3.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.k_cache: torch.Size([1, 8, 96, 128]) - layer.4.v_cache: torch.Size([1, 8, 96, 128]) - layer.4.output: torch.Size([1, 96, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02591230 143.48148600 - layer.0.v_cache 0.00000028 0.00067859 - layer.1.k_cache 0.00321835 7.00868225 - layer.1.v_cache 0.00000081 0.00283153 - layer.2.k_cache 0.00115341 2.27547868 - layer.2.v_cache 0.00000115 0.00423899 - layer.3.k_cache 0.00138578 2.72529221 - layer.3.v_cache 0.00000239 0.00770842 - layer.4.k_cache 0.00338578 6.82825216 - layer.4.v_cache 0.00000322 0.01224912 - layer.4.output 0.00018605 0.25391408 - ------------------------------------------------------------------------------------- - TOTAL 0.00255769 11.66875388 - (elements=1,376,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1376256 -Total Bytes 17068 -BPFP 0.0992 bits/point -EBPFP 0.1984 equivalent bits/point -MSE 11.668754 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.005s, Pack+Encode: 0.204s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 96, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 96, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.6688 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample227-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample227-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample233-layer4-item1.zst (35/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample233-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 85, 128) -Output shape: (1, 85, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.output: torch.Size([1, 85, 4096]) -> torch.Size([1, 1, 85, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 928B, BPFP=0.0853 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 968B, BPFP=0.0890 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,336B, BPFP=0.1228 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,156B, BPFP=0.1062 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,412B, BPFP=0.1298 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,220B, BPFP=0.1121 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,468B, BPFP=0.1349 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,304B, BPFP=0.1199 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,220B, BPFP=0.1121 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,248B, BPFP=0.1147 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,676B, BPFP=0.0845 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.278s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02725663 147.54515165 - layer.0.v_cache 0.00000027 0.00068413 - layer.1.k_cache 0.00328410 7.08623262 - layer.1.v_cache 0.00000085 0.00284120 - layer.2.k_cache 0.00114159 2.23306256 - layer.2.v_cache 0.00000125 0.00448724 - layer.3.k_cache 0.00133976 2.77242665 - layer.3.v_cache 0.00000266 0.00775524 - layer.4.k_cache 0.00332377 6.71104952 - layer.4.v_cache 0.00000336 0.01246122 - layer.4.output 0.00023926 0.26379446 - ------------------------------------------------------------------------------------- - TOTAL 0.00266509 11.95938071 - (elements=1,218,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1218560 -Total Bytes 15936 -BPFP 0.1046 bits/point -EBPFP 0.2092 equivalent bits/point -MSE 11.959381 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.278s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.9594 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample233-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample233-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample241-layer4-item1.zst (36/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample241-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 948B, BPFP=0.0764 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,020B, BPFP=0.0822 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,376B, BPFP=0.1108 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,204B, BPFP=0.0970 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,524B, BPFP=0.1227 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,320B, BPFP=0.1063 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,548B, BPFP=0.1247 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,396B, BPFP=0.1124 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,376B, BPFP=0.1108 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,360B, BPFP=0.1095 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,720B, BPFP=0.0749 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02553191 132.74699984 - layer.0.v_cache 0.00000029 0.00068947 - layer.1.k_cache 0.00322990 6.85820212 - layer.1.v_cache 0.00000089 0.00273460 - layer.2.k_cache 0.00117012 2.26263506 - layer.2.v_cache 0.00000108 0.00415163 - layer.3.k_cache 0.00135684 2.68913883 - layer.3.v_cache 0.00000209 0.00700112 - layer.4.k_cache 0.00331928 6.68066406 - layer.4.v_cache 0.00000299 0.01148881 - layer.4.output 0.00021425 0.26124126 - ------------------------------------------------------------------------------------- - TOTAL 0.00253374 10.87919076 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 16792 -BPFP 0.0966 bits/point -EBPFP 0.1932 equivalent bits/point -MSE 10.879191 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.005s, Pack+Encode: 0.203s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.8792 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample241-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample241-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample250-layer4-item1.zst (37/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample250-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 97, 128) -Output shape: (1, 97, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) -> torch.Size([1, 1, 97, 1024]) - layer.4.output: torch.Size([1, 97, 4096]) -> torch.Size([1, 1, 97, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 944B, BPFP=0.0760 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,000B, BPFP=0.0805 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,380B, BPFP=0.1111 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,256B, BPFP=0.1012 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,532B, BPFP=0.1234 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,372B, BPFP=0.1105 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,520B, BPFP=0.1224 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,428B, BPFP=0.1150 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,348B, BPFP=0.1086 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,376B, BPFP=0.1108 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,968B, BPFP=0.0799 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 97, 128]) - layer.0.v_cache: torch.Size([1, 8, 97, 128]) - layer.1.k_cache: torch.Size([1, 8, 97, 128]) - layer.1.v_cache: torch.Size([1, 8, 97, 128]) - layer.2.k_cache: torch.Size([1, 8, 97, 128]) - layer.2.v_cache: torch.Size([1, 8, 97, 128]) - layer.3.k_cache: torch.Size([1, 8, 97, 128]) - layer.3.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.k_cache: torch.Size([1, 8, 97, 128]) - layer.4.v_cache: torch.Size([1, 8, 97, 128]) - layer.4.output: torch.Size([1, 97, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02408928 131.36863724 - layer.0.v_cache 0.00000028 0.00069699 - layer.1.k_cache 0.00330002 6.74358690 - layer.1.v_cache 0.00000090 0.00276439 - layer.2.k_cache 0.00113900 2.21114562 - layer.2.v_cache 0.00000130 0.00429979 - layer.3.k_cache 0.00136322 2.72211732 - layer.3.v_cache 0.00000251 0.00787142 - layer.4.k_cache 0.00342626 6.72127321 - layer.4.v_cache 0.00000319 0.01213876 - layer.4.output 0.00021399 0.26900995 - ------------------------------------------------------------------------------------- - TOTAL 0.00244156 10.77646939 - (elements=1,390,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1390592 -Total Bytes 17124 -BPFP 0.0985 bits/point -EBPFP 0.1970 equivalent bits/point -MSE 10.776469 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 97, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 97, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.7765 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample250-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample250-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample251-layer4-item1.zst (38/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample251-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 99, 128) -Output shape: (1, 99, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) -> torch.Size([1, 1, 99, 1024]) - layer.4.output: torch.Size([1, 99, 4096]) -> torch.Size([1, 1, 99, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 936B, BPFP=0.0739 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,080B, BPFP=0.0852 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,420B, BPFP=0.1121 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,200B, BPFP=0.0947 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,524B, BPFP=0.1203 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,340B, BPFP=0.1057 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,536B, BPFP=0.1212 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,432B, BPFP=0.1130 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,372B, BPFP=0.1083 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,420B, BPFP=0.1121 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,936B, BPFP=0.0777 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.output: torch.Size([1, 99, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 99, 128]) - layer.0.v_cache: torch.Size([1, 8, 99, 128]) - layer.1.k_cache: torch.Size([1, 8, 99, 128]) - layer.1.v_cache: torch.Size([1, 8, 99, 128]) - layer.2.k_cache: torch.Size([1, 8, 99, 128]) - layer.2.v_cache: torch.Size([1, 8, 99, 128]) - layer.3.k_cache: torch.Size([1, 8, 99, 128]) - layer.3.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.k_cache: torch.Size([1, 8, 99, 128]) - layer.4.v_cache: torch.Size([1, 8, 99, 128]) - layer.4.output: torch.Size([1, 99, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02490362 126.24071772 - layer.0.v_cache 0.00000029 0.00068092 - layer.1.k_cache 0.00335574 6.83505342 - layer.1.v_cache 0.00000081 0.00249499 - layer.2.k_cache 0.00112037 2.23902431 - layer.2.v_cache 0.00000110 0.00392598 - layer.3.k_cache 0.00138480 2.73110808 - layer.3.v_cache 0.00000194 0.00662826 - layer.4.k_cache 0.00346640 6.90502930 - layer.4.v_cache 0.00000296 0.01069472 - layer.4.output 0.00019387 0.24646780 - ------------------------------------------------------------------------------------- - TOTAL 0.00250096 10.42580206 - (elements=1,419,264) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1419264 -Total Bytes 17196 -BPFP 0.0969 bits/point -EBPFP 0.1939 equivalent bits/point -MSE 10.425802 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.006s, Pack+Encode: 0.203s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 99, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 99, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.4258 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample251-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample251-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample257-layer4-item1.zst (39/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample257-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 86, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 86, 128) -Output shape: (1, 86, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.0.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.1.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.1.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.2.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.2.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.3.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.3.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.4.k_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.4.v_cache: torch.Size([1, 8, 86, 128]) -> torch.Size([1, 1, 86, 1024]) - layer.4.output: torch.Size([1, 86, 4096]) -> torch.Size([1, 1, 86, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 948B, BPFP=0.0861 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,068B, BPFP=0.0970 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,368B, BPFP=0.1243 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,204B, BPFP=0.1094 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,436B, BPFP=0.1305 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,368B, BPFP=0.1243 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,476B, BPFP=0.1341 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,380B, BPFP=0.1254 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,260B, BPFP=0.1145 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,320B, BPFP=0.1199 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,984B, BPFP=0.0905 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 86, 128]) - layer.0.v_cache: torch.Size([1, 8, 86, 128]) - layer.1.k_cache: torch.Size([1, 8, 86, 128]) - layer.1.v_cache: torch.Size([1, 8, 86, 128]) - layer.2.k_cache: torch.Size([1, 8, 86, 128]) - layer.2.v_cache: torch.Size([1, 8, 86, 128]) - layer.3.k_cache: torch.Size([1, 8, 86, 128]) - layer.3.v_cache: torch.Size([1, 8, 86, 128]) - layer.4.k_cache: torch.Size([1, 8, 86, 128]) - layer.4.v_cache: torch.Size([1, 8, 86, 128]) - layer.4.output: torch.Size([1, 86, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 86, 128]) - layer.0.v_cache: torch.Size([1, 8, 86, 128]) - layer.1.k_cache: torch.Size([1, 8, 86, 128]) - layer.1.v_cache: torch.Size([1, 8, 86, 128]) - layer.2.k_cache: torch.Size([1, 8, 86, 128]) - layer.2.v_cache: torch.Size([1, 8, 86, 128]) - layer.3.k_cache: torch.Size([1, 8, 86, 128]) - layer.3.v_cache: torch.Size([1, 8, 86, 128]) - layer.4.k_cache: torch.Size([1, 8, 86, 128]) - layer.4.v_cache: torch.Size([1, 8, 86, 128]) - layer.4.output: torch.Size([1, 86, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02609516 167.58493823 - layer.0.v_cache 0.00000027 0.00067943 - layer.1.k_cache 0.00348465 7.10743323 - layer.1.v_cache 0.00000080 0.00281071 - layer.2.k_cache 0.00118276 2.25951173 - layer.2.v_cache 0.00000133 0.00477609 - layer.3.k_cache 0.00137135 2.77372600 - layer.3.v_cache 0.00000258 0.00785011 - layer.4.k_cache 0.00338397 6.88388842 - layer.4.v_cache 0.00000307 0.01182296 - layer.4.output 0.00019889 0.25908459 - ------------------------------------------------------------------------------------- - TOTAL 0.00259439 13.40526966 - (elements=1,232,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1232896 -Total Bytes 16812 -BPFP 0.1091 bits/point -EBPFP 0.2182 equivalent bits/point -MSE 13.405270 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 86, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 86, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 13.4053 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample257-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample257-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample258-layer4-item1.zst (40/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample258-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 89, 128) -Output shape: (1, 89, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.output: torch.Size([1, 89, 4096]) -> torch.Size([1, 1, 89, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 984B, BPFP=0.0864 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,028B, BPFP=0.0902 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,420B, BPFP=0.1246 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,168B, BPFP=0.1025 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,492B, BPFP=0.1310 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,268B, BPFP=0.1113 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,536B, BPFP=0.1348 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,356B, BPFP=0.1190 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,344B, BPFP=0.1180 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,304B, BPFP=0.1145 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,692B, BPFP=0.0810 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02681794 133.86731917 - layer.0.v_cache 0.00000027 0.00067696 - layer.1.k_cache 0.00344681 6.71187429 - layer.1.v_cache 0.00000079 0.00271128 - layer.2.k_cache 0.00114729 2.27478370 - layer.2.v_cache 0.00000121 0.00416870 - layer.3.k_cache 0.00138721 2.71867628 - layer.3.v_cache 0.00000216 0.00732979 - layer.4.k_cache 0.00341441 6.80345720 - layer.4.v_cache 0.00000316 0.01190262 - layer.4.output 0.00022172 0.27734156 - ------------------------------------------------------------------------------------- - TOTAL 0.00265058 10.96516187 - (elements=1,275,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1275904 -Total Bytes 16592 -BPFP 0.1040 bits/point -EBPFP 0.2081 equivalent bits/point -MSE 10.965162 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.9652 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample258-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample258-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample263-layer4-item1.zst (41/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample263-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 89, 128) -Output shape: (1, 89, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) -> torch.Size([1, 1, 89, 1024]) - layer.4.output: torch.Size([1, 89, 4096]) -> torch.Size([1, 1, 89, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 992B, BPFP=0.0871 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,112B, BPFP=0.0976 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,432B, BPFP=0.1257 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,284B, BPFP=0.1127 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,492B, BPFP=0.1310 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,352B, BPFP=0.1187 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,564B, BPFP=0.1373 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,420B, BPFP=0.1246 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,372B, BPFP=0.1204 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,372B, BPFP=0.1204 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,920B, BPFP=0.0860 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 89, 128]) - layer.0.v_cache: torch.Size([1, 8, 89, 128]) - layer.1.k_cache: torch.Size([1, 8, 89, 128]) - layer.1.v_cache: torch.Size([1, 8, 89, 128]) - layer.2.k_cache: torch.Size([1, 8, 89, 128]) - layer.2.v_cache: torch.Size([1, 8, 89, 128]) - layer.3.k_cache: torch.Size([1, 8, 89, 128]) - layer.3.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.k_cache: torch.Size([1, 8, 89, 128]) - layer.4.v_cache: torch.Size([1, 8, 89, 128]) - layer.4.output: torch.Size([1, 89, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02424761 138.32854196 - layer.0.v_cache 0.00000029 0.00069286 - layer.1.k_cache 0.00343373 6.85473701 - layer.1.v_cache 0.00000078 0.00281847 - layer.2.k_cache 0.00114421 2.21768737 - layer.2.v_cache 0.00000115 0.00435065 - layer.3.k_cache 0.00135741 2.72249338 - layer.3.v_cache 0.00000236 0.00757348 - layer.4.k_cache 0.00332665 6.61865852 - layer.4.v_cache 0.00000319 0.01195603 - layer.4.output 0.00018046 0.24424787 - ------------------------------------------------------------------------------------- - TOTAL 0.00244566 11.26760723 - (elements=1,275,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1275904 -Total Bytes 17312 -BPFP 0.1085 bits/point -EBPFP 0.2171 equivalent bits/point -MSE 11.267607 ----------------------- -------------------------------------------------------- -Time: 0.488s Load: 0.006s, Pack+Encode: 0.201s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 89, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 89, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.2676 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample263-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample263-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample274-layer4-item1.zst (42/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample274-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 92, 128) -Output shape: (1, 92, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) -> torch.Size([1, 1, 92, 1024]) - layer.4.output: torch.Size([1, 92, 4096]) -> torch.Size([1, 1, 92, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 988B, BPFP=0.0839 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,040B, BPFP=0.0883 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,364B, BPFP=0.1158 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,148B, BPFP=0.0975 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,456B, BPFP=0.1236 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,336B, BPFP=0.1135 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,528B, BPFP=0.1298 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,352B, BPFP=0.1148 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,360B, BPFP=0.1155 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,356B, BPFP=0.1151 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,656B, BPFP=0.0776 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 92, 128]) - layer.0.v_cache: torch.Size([1, 8, 92, 128]) - layer.1.k_cache: torch.Size([1, 8, 92, 128]) - layer.1.v_cache: torch.Size([1, 8, 92, 128]) - layer.2.k_cache: torch.Size([1, 8, 92, 128]) - layer.2.v_cache: torch.Size([1, 8, 92, 128]) - layer.3.k_cache: torch.Size([1, 8, 92, 128]) - layer.3.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.k_cache: torch.Size([1, 8, 92, 128]) - layer.4.v_cache: torch.Size([1, 8, 92, 128]) - layer.4.output: torch.Size([1, 92, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02551763 83.55144468 - layer.0.v_cache 0.00000027 0.00065108 - layer.1.k_cache 0.00347722 6.54707668 - layer.1.v_cache 0.00000076 0.00243644 - layer.2.k_cache 0.00117662 2.18612256 - layer.2.v_cache 0.00000108 0.00385678 - layer.3.k_cache 0.00141157 2.64213363 - layer.3.v_cache 0.00000195 0.00650242 - layer.4.k_cache 0.00338827 6.57521654 - layer.4.v_cache 0.00000294 0.01103471 - layer.4.output 0.00021021 0.24682271 - ------------------------------------------------------------------------------------- - TOTAL 0.00255851 7.32241188 - (elements=1,318,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1318912 -Total Bytes 16584 -BPFP 0.1006 bits/point -EBPFP 0.2012 equivalent bits/point -MSE 7.322412 ----------------------- -------------------------------------------------------- -Time: 0.488s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 92, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 92, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.3224 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample274-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample274-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample282-layer4-item1.zst (43/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample282-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 87, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 87, 128) -Output shape: (1, 87, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.0.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.1.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.1.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.2.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.2.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.3.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.3.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.4.k_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.4.v_cache: torch.Size([1, 8, 87, 128]) -> torch.Size([1, 1, 87, 1024]) - layer.4.output: torch.Size([1, 87, 4096]) -> torch.Size([1, 1, 87, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 964B, BPFP=0.0866 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,024B, BPFP=0.0920 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,400B, BPFP=0.1257 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,180B, BPFP=0.1060 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,476B, BPFP=0.1325 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,320B, BPFP=0.1185 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,512B, BPFP=0.1358 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,380B, BPFP=0.1239 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,280B, BPFP=0.1149 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,312B, BPFP=0.1178 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,908B, BPFP=0.0877 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 87, 128]) - layer.0.v_cache: torch.Size([1, 8, 87, 128]) - layer.1.k_cache: torch.Size([1, 8, 87, 128]) - layer.1.v_cache: torch.Size([1, 8, 87, 128]) - layer.2.k_cache: torch.Size([1, 8, 87, 128]) - layer.2.v_cache: torch.Size([1, 8, 87, 128]) - layer.3.k_cache: torch.Size([1, 8, 87, 128]) - layer.3.v_cache: torch.Size([1, 8, 87, 128]) - layer.4.k_cache: torch.Size([1, 8, 87, 128]) - layer.4.v_cache: torch.Size([1, 8, 87, 128]) - layer.4.output: torch.Size([1, 87, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 87, 128]) - layer.0.v_cache: torch.Size([1, 8, 87, 128]) - layer.1.k_cache: torch.Size([1, 8, 87, 128]) - layer.1.v_cache: torch.Size([1, 8, 87, 128]) - layer.2.k_cache: torch.Size([1, 8, 87, 128]) - layer.2.v_cache: torch.Size([1, 8, 87, 128]) - layer.3.k_cache: torch.Size([1, 8, 87, 128]) - layer.3.v_cache: torch.Size([1, 8, 87, 128]) - layer.4.k_cache: torch.Size([1, 8, 87, 128]) - layer.4.v_cache: torch.Size([1, 8, 87, 128]) - layer.4.output: torch.Size([1, 87, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02591714 134.40783270 - layer.0.v_cache 0.00000026 0.00067305 - layer.1.k_cache 0.00374749 7.04239348 - layer.1.v_cache 0.00000077 0.00262839 - layer.2.k_cache 0.00116905 2.25849581 - layer.2.v_cache 0.00000114 0.00409304 - layer.3.k_cache 0.00135864 2.69338446 - layer.3.v_cache 0.00000233 0.00726719 - layer.4.k_cache 0.00339113 6.69832357 - layer.4.v_cache 0.00000304 0.01123666 - layer.4.output 0.00030438 0.25667978 - ------------------------------------------------------------------------------------- - TOTAL 0.00262918 11.01093196 - (elements=1,247,232) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1247232 -Total Bytes 16756 -BPFP 0.1075 bits/point -EBPFP 0.2150 equivalent bits/point -MSE 11.010932 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.005s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 87, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 87, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.0109 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample282-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample282-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample290-layer4-item1.zst (44/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample290-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 94, 128) -Output shape: (1, 94, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.output: torch.Size([1, 94, 4096]) -> torch.Size([1, 1, 94, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 992B, BPFP=0.0824 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,052B, BPFP=0.0874 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,364B, BPFP=0.1134 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,216B, BPFP=0.1011 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,540B, BPFP=0.1280 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,344B, BPFP=0.1117 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,516B, BPFP=0.1260 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,532B, BPFP=0.1273 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,380B, BPFP=0.1147 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,448B, BPFP=0.1203 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,956B, BPFP=0.0822 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02423150 88.66771941 - layer.0.v_cache 0.00000028 0.00065305 - layer.1.k_cache 0.00344140 6.60343771 - layer.1.v_cache 0.00000074 0.00252918 - layer.2.k_cache 0.00110852 2.09966359 - layer.2.v_cache 0.00000113 0.00398517 - layer.3.k_cache 0.00135732 2.63745344 - layer.3.v_cache 0.00000223 0.00716051 - layer.4.k_cache 0.00337875 6.49124340 - layer.4.v_cache 0.00000288 0.01081061 - layer.4.output 0.00018208 0.23287838 - ------------------------------------------------------------------------------------- - TOTAL 0.00244665 7.67544069 - (elements=1,347,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1347584 -Total Bytes 17340 -BPFP 0.1029 bits/point -EBPFP 0.2059 equivalent bits/point -MSE 7.675441 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.6754 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample290-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample290-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample3-layer4-item1.zst (45/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample3-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 149, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 149, 128) -Output shape: (1, 149, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.0.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.1.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.1.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.2.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.2.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.3.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.3.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.4.k_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.4.v_cache: torch.Size([1, 8, 149, 128]) -> torch.Size([1, 1, 149, 1024]) - layer.4.output: torch.Size([1, 149, 4096]) -> torch.Size([1, 1, 149, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,584B, BPFP=0.0831 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,672B, BPFP=0.0877 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,176B, BPFP=0.1141 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,908B, BPFP=0.1000 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,288B, BPFP=0.1200 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,004B, BPFP=0.1051 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,344B, BPFP=0.1229 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,104B, BPFP=0.1103 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,072B, BPFP=0.1086 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,068B, BPFP=0.1084 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 6,392B, BPFP=0.0838 -⌛️ [2/4] FRONTEND: Frontend time: 0.251s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 149, 128]) - layer.0.v_cache: torch.Size([1, 8, 149, 128]) - layer.1.k_cache: torch.Size([1, 8, 149, 128]) - layer.1.v_cache: torch.Size([1, 8, 149, 128]) - layer.2.k_cache: torch.Size([1, 8, 149, 128]) - layer.2.v_cache: torch.Size([1, 8, 149, 128]) - layer.3.k_cache: torch.Size([1, 8, 149, 128]) - layer.3.v_cache: torch.Size([1, 8, 149, 128]) - layer.4.k_cache: torch.Size([1, 8, 149, 128]) - layer.4.v_cache: torch.Size([1, 8, 149, 128]) - layer.4.output: torch.Size([1, 149, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.376s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 149, 128]) - layer.0.v_cache: torch.Size([1, 8, 149, 128]) - layer.1.k_cache: torch.Size([1, 8, 149, 128]) - layer.1.v_cache: torch.Size([1, 8, 149, 128]) - layer.2.k_cache: torch.Size([1, 8, 149, 128]) - layer.2.v_cache: torch.Size([1, 8, 149, 128]) - layer.3.k_cache: torch.Size([1, 8, 149, 128]) - layer.3.v_cache: torch.Size([1, 8, 149, 128]) - layer.4.k_cache: torch.Size([1, 8, 149, 128]) - layer.4.v_cache: torch.Size([1, 8, 149, 128]) - layer.4.output: torch.Size([1, 149, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02473939 94.31977506 - layer.0.v_cache 0.00000028 0.00068456 - layer.1.k_cache 0.00314236 6.30376132 - layer.1.v_cache 0.00000089 0.00277387 - layer.2.k_cache 0.00115430 2.11570750 - layer.2.v_cache 0.00000109 0.00415443 - layer.3.k_cache 0.00134690 2.56707887 - layer.3.v_cache 0.00000207 0.00713977 - layer.4.k_cache 0.00351049 6.64964469 - layer.4.v_cache 0.00000318 0.01198229 - layer.4.output 0.00015300 0.21119512 - ------------------------------------------------------------------------------------- - TOTAL 0.00246521 8.05910592 - (elements=2,136,064) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2136064 -Total Bytes 26612 -BPFP 0.0997 bits/point -EBPFP 0.1993 equivalent bits/point -MSE 8.059106 ----------------------- -------------------------------------------------------- -Time: 0.635s Load: 0.009s, Pack+Encode: 0.251s, Decode+Unpack: 0.376s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 149, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 149, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.0591 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample3-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample3-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample307-layer4-item1.zst (46/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample307-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 93, 128) -Output shape: (1, 93, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.output: torch.Size([1, 93, 4096]) -> torch.Size([1, 1, 93, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,004B, BPFP=0.0843 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,112B, BPFP=0.0934 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,396B, BPFP=0.1173 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,304B, BPFP=0.1095 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,528B, BPFP=0.1284 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,404B, BPFP=0.1179 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,592B, BPFP=0.1337 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,504B, BPFP=0.1263 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,380B, BPFP=0.1159 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,448B, BPFP=0.1216 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,924B, BPFP=0.0824 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02417885 84.17692057 - layer.0.v_cache 0.00000029 0.00068301 - layer.1.k_cache 0.00337012 6.40353820 - layer.1.v_cache 0.00000087 0.00266845 - layer.2.k_cache 0.00111931 2.15172716 - layer.2.v_cache 0.00000117 0.00405509 - layer.3.k_cache 0.00133665 2.66333303 - layer.3.v_cache 0.00000243 0.00726542 - layer.4.k_cache 0.00331438 6.50943420 - layer.4.v_cache 0.00000316 0.01168931 - layer.4.output 0.00021472 0.25363391 - ------------------------------------------------------------------------------------- - TOTAL 0.00244186 7.35327501 - (elements=1,333,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1333248 -Total Bytes 17596 -BPFP 0.1056 bits/point -EBPFP 0.2112 equivalent bits/point -MSE 7.353275 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.005s, Pack+Encode: 0.203s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.3533 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample307-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample307-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample313-layer4-item1.zst (47/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample313-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 81, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 81, 128) -Output shape: (1, 81, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.output: torch.Size([1, 81, 4096]) -> torch.Size([1, 1, 81, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 916B, BPFP=0.0883 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 948B, BPFP=0.0914 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,312B, BPFP=0.1265 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,108B, BPFP=0.1069 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,404B, BPFP=0.1354 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,216B, BPFP=0.1173 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,436B, BPFP=0.1385 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,276B, BPFP=0.1231 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,192B, BPFP=0.1150 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,196B, BPFP=0.1154 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,732B, BPFP=0.0900 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.output: torch.Size([1, 81, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.output: torch.Size([1, 81, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02598516 88.19507137 - layer.0.v_cache 0.00000026 0.00065397 - layer.1.k_cache 0.00346748 7.09286311 - layer.1.v_cache 0.00000092 0.00270536 - layer.2.k_cache 0.00113221 2.25031026 - layer.2.v_cache 0.00000143 0.00430413 - layer.3.k_cache 0.00133539 2.72006263 - layer.3.v_cache 0.00000229 0.00730261 - layer.4.k_cache 0.00342405 6.58903145 - layer.4.v_cache 0.00000310 0.01162734 - layer.4.output 0.00020696 0.24694188 - ------------------------------------------------------------------------------------- - TOTAL 0.00258429 7.70440713 - (elements=1,161,216) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1161216 -Total Bytes 15736 -BPFP 0.1084 bits/point -EBPFP 0.2168 equivalent bits/point -MSE 7.704407 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.006s, Pack+Encode: 0.204s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 81, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.7044 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample313-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample313-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample319-layer4-item1.zst (48/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample319-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 93, 128) -Output shape: (1, 93, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) -> torch.Size([1, 1, 93, 1024]) - layer.4.output: torch.Size([1, 93, 4096]) -> torch.Size([1, 1, 93, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 976B, BPFP=0.0820 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,064B, BPFP=0.0894 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,416B, BPFP=0.1190 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,236B, BPFP=0.1038 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,496B, BPFP=0.1257 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,376B, BPFP=0.1156 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,524B, BPFP=0.1280 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,380B, BPFP=0.1159 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,372B, BPFP=0.1153 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,388B, BPFP=0.1166 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,836B, BPFP=0.0806 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 93, 128]) - layer.0.v_cache: torch.Size([1, 8, 93, 128]) - layer.1.k_cache: torch.Size([1, 8, 93, 128]) - layer.1.v_cache: torch.Size([1, 8, 93, 128]) - layer.2.k_cache: torch.Size([1, 8, 93, 128]) - layer.2.v_cache: torch.Size([1, 8, 93, 128]) - layer.3.k_cache: torch.Size([1, 8, 93, 128]) - layer.3.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.k_cache: torch.Size([1, 8, 93, 128]) - layer.4.v_cache: torch.Size([1, 8, 93, 128]) - layer.4.output: torch.Size([1, 93, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02450277 81.27656670 - layer.0.v_cache 0.00000028 0.00066344 - layer.1.k_cache 0.00336978 6.70506237 - layer.1.v_cache 0.00000078 0.00263613 - layer.2.k_cache 0.00117610 2.16850035 - layer.2.v_cache 0.00000111 0.00400532 - layer.3.k_cache 0.00133121 2.65510641 - layer.3.v_cache 0.00000213 0.00687363 - layer.4.k_cache 0.00346230 6.60170820 - layer.4.v_cache 0.00000315 0.01120994 - layer.4.output 0.00018648 0.23623626 - ------------------------------------------------------------------------------------- - TOTAL 0.00247111 7.16980554 - (elements=1,333,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1333248 -Total Bytes 17064 -BPFP 0.1024 bits/point -EBPFP 0.2048 equivalent bits/point -MSE 7.169806 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.007s, Pack+Encode: 0.202s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 93, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 93, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.1698 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample319-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample319-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample333-layer4-item1.zst (49/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample333-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 90, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 90, 128) -Output shape: (1, 90, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.0.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.1.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.1.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.2.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.2.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.3.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.3.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.4.k_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.4.v_cache: torch.Size([1, 8, 90, 128]) -> torch.Size([1, 1, 90, 1024]) - layer.4.output: torch.Size([1, 90, 4096]) -> torch.Size([1, 1, 90, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 996B, BPFP=0.0865 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,072B, BPFP=0.0931 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,396B, BPFP=0.1212 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,260B, BPFP=0.1094 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,468B, BPFP=0.1274 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,332B, BPFP=0.1156 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,536B, BPFP=0.1333 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,384B, BPFP=0.1201 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,352B, BPFP=0.1174 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,352B, BPFP=0.1174 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,744B, BPFP=0.0813 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 90, 128]) - layer.0.v_cache: torch.Size([1, 8, 90, 128]) - layer.1.k_cache: torch.Size([1, 8, 90, 128]) - layer.1.v_cache: torch.Size([1, 8, 90, 128]) - layer.2.k_cache: torch.Size([1, 8, 90, 128]) - layer.2.v_cache: torch.Size([1, 8, 90, 128]) - layer.3.k_cache: torch.Size([1, 8, 90, 128]) - layer.3.v_cache: torch.Size([1, 8, 90, 128]) - layer.4.k_cache: torch.Size([1, 8, 90, 128]) - layer.4.v_cache: torch.Size([1, 8, 90, 128]) - layer.4.output: torch.Size([1, 90, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 90, 128]) - layer.0.v_cache: torch.Size([1, 8, 90, 128]) - layer.1.k_cache: torch.Size([1, 8, 90, 128]) - layer.1.v_cache: torch.Size([1, 8, 90, 128]) - layer.2.k_cache: torch.Size([1, 8, 90, 128]) - layer.2.v_cache: torch.Size([1, 8, 90, 128]) - layer.3.k_cache: torch.Size([1, 8, 90, 128]) - layer.3.v_cache: torch.Size([1, 8, 90, 128]) - layer.4.k_cache: torch.Size([1, 8, 90, 128]) - layer.4.v_cache: torch.Size([1, 8, 90, 128]) - layer.4.output: torch.Size([1, 90, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02906363 134.52450087 - layer.0.v_cache 0.00000028 0.00069791 - layer.1.k_cache 0.00333766 6.82069431 - layer.1.v_cache 0.00000075 0.00253210 - layer.2.k_cache 0.00114903 2.19759894 - layer.2.v_cache 0.00000112 0.00396400 - layer.3.k_cache 0.00131913 2.71781684 - layer.3.v_cache 0.00000216 0.00705144 - layer.4.k_cache 0.00329656 6.69507243 - layer.4.v_cache 0.00000300 0.01109646 - layer.4.output 0.00024019 0.25577789 - ------------------------------------------------------------------------------------- - TOTAL 0.00279529 11.00029549 - (elements=1,290,240) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1290240 -Total Bytes 16892 -BPFP 0.1047 bits/point -EBPFP 0.2095 equivalent bits/point -MSE 11.000295 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.005s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 90, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 90, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.0003 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample333-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample333-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample36-layer4-item1.zst (50/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample36-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 119, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 119, 128) -Output shape: (1, 119, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.0.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.1.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.1.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.2.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.2.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.3.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.3.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.4.k_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.4.v_cache: torch.Size([1, 8, 119, 128]) -> torch.Size([1, 1, 119, 1024]) - layer.4.output: torch.Size([1, 119, 4096]) -> torch.Size([1, 1, 119, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,116B, BPFP=0.0733 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,100B, BPFP=0.0722 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,524B, BPFP=0.1001 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,356B, BPFP=0.0890 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,620B, BPFP=0.1064 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,448B, BPFP=0.0951 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,652B, BPFP=0.1085 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,524B, BPFP=0.1001 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,508B, BPFP=0.0990 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,468B, BPFP=0.0964 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,980B, BPFP=0.0653 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 119, 128]) - layer.0.v_cache: torch.Size([1, 8, 119, 128]) - layer.1.k_cache: torch.Size([1, 8, 119, 128]) - layer.1.v_cache: torch.Size([1, 8, 119, 128]) - layer.2.k_cache: torch.Size([1, 8, 119, 128]) - layer.2.v_cache: torch.Size([1, 8, 119, 128]) - layer.3.k_cache: torch.Size([1, 8, 119, 128]) - layer.3.v_cache: torch.Size([1, 8, 119, 128]) - layer.4.k_cache: torch.Size([1, 8, 119, 128]) - layer.4.v_cache: torch.Size([1, 8, 119, 128]) - layer.4.output: torch.Size([1, 119, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 119, 128]) - layer.0.v_cache: torch.Size([1, 8, 119, 128]) - layer.1.k_cache: torch.Size([1, 8, 119, 128]) - layer.1.v_cache: torch.Size([1, 8, 119, 128]) - layer.2.k_cache: torch.Size([1, 8, 119, 128]) - layer.2.v_cache: torch.Size([1, 8, 119, 128]) - layer.3.k_cache: torch.Size([1, 8, 119, 128]) - layer.3.v_cache: torch.Size([1, 8, 119, 128]) - layer.4.k_cache: torch.Size([1, 8, 119, 128]) - layer.4.v_cache: torch.Size([1, 8, 119, 128]) - layer.4.output: torch.Size([1, 119, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02458241 203.13499540 - layer.0.v_cache 0.00000028 0.00069835 - layer.1.k_cache 0.00331476 7.38343208 - layer.1.v_cache 0.00000080 0.00290362 - layer.2.k_cache 0.00114230 2.34445601 - layer.2.v_cache 0.00000115 0.00456068 - layer.3.k_cache 0.00138709 2.84642555 - layer.3.v_cache 0.00000225 0.00780249 - layer.4.k_cache 0.00333808 7.45315834 - layer.4.v_cache 0.00000314 0.01221402 - layer.4.output 0.00020645 0.27213710 - ------------------------------------------------------------------------------------- - TOTAL 0.00247129 16.01994250 - (elements=1,705,984) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1705984 -Total Bytes 18296 -BPFP 0.0858 bits/point -EBPFP 0.1716 equivalent bits/point -MSE 16.019942 ----------------------- -------------------------------------------------------- -Time: 0.495s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 119, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 119, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 16.0199 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample36-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample36-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample365-layer4-item1.zst (51/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample365-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 82, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 82, 128) -Output shape: (1, 82, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) -> torch.Size([1, 1, 82, 1024]) - layer.4.output: torch.Size([1, 82, 4096]) -> torch.Size([1, 1, 82, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 904B, BPFP=0.0861 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 968B, BPFP=0.0922 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,296B, BPFP=0.1235 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,112B, BPFP=0.1059 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,408B, BPFP=0.1341 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,216B, BPFP=0.1159 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1372 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,272B, BPFP=0.1212 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,200B, BPFP=0.1143 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,200B, BPFP=0.1143 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,552B, BPFP=0.0846 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.output: torch.Size([1, 82, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 82, 128]) - layer.0.v_cache: torch.Size([1, 8, 82, 128]) - layer.1.k_cache: torch.Size([1, 8, 82, 128]) - layer.1.v_cache: torch.Size([1, 8, 82, 128]) - layer.2.k_cache: torch.Size([1, 8, 82, 128]) - layer.2.v_cache: torch.Size([1, 8, 82, 128]) - layer.3.k_cache: torch.Size([1, 8, 82, 128]) - layer.3.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.k_cache: torch.Size([1, 8, 82, 128]) - layer.4.v_cache: torch.Size([1, 8, 82, 128]) - layer.4.output: torch.Size([1, 82, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02610582 89.20600110 - layer.0.v_cache 0.00000027 0.00066468 - layer.1.k_cache 0.00344214 6.58516935 - layer.1.v_cache 0.00000078 0.00261697 - layer.2.k_cache 0.00114699 2.21638563 - layer.2.v_cache 0.00000119 0.00417543 - layer.3.k_cache 0.00140111 2.73139172 - layer.3.v_cache 0.00000200 0.00691752 - layer.4.k_cache 0.00332288 6.58971051 - layer.4.v_cache 0.00000299 0.01074026 - layer.4.output 0.00018752 0.24995383 - ------------------------------------------------------------------------------------- - TOTAL 0.00258402 7.73954204 - (elements=1,175,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1175552 -Total Bytes 15568 -BPFP 0.1059 bits/point -EBPFP 0.2119 equivalent bits/point -MSE 7.739542 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.005s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 82, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 82, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.7395 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample365-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample365-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample38-layer4-item1.zst (52/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample38-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 124, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 124, 128) -Output shape: (1, 124, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.0.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.1.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.1.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.2.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.2.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.3.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.3.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.4.k_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.4.v_cache: torch.Size([1, 8, 124, 128]) -> torch.Size([1, 1, 124, 1024]) - layer.4.output: torch.Size([1, 124, 4096]) -> torch.Size([1, 1, 124, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,028B, BPFP=0.0648 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,072B, BPFP=0.0675 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,304B, BPFP=0.0822 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,148B, BPFP=0.0723 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,384B, BPFP=0.0872 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,208B, BPFP=0.0761 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,424B, BPFP=0.0897 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,240B, BPFP=0.0781 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,268B, BPFP=0.0799 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,220B, BPFP=0.0769 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,528B, BPFP=0.0556 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 124, 128]) - layer.0.v_cache: torch.Size([1, 8, 124, 128]) - layer.1.k_cache: torch.Size([1, 8, 124, 128]) - layer.1.v_cache: torch.Size([1, 8, 124, 128]) - layer.2.k_cache: torch.Size([1, 8, 124, 128]) - layer.2.v_cache: torch.Size([1, 8, 124, 128]) - layer.3.k_cache: torch.Size([1, 8, 124, 128]) - layer.3.v_cache: torch.Size([1, 8, 124, 128]) - layer.4.k_cache: torch.Size([1, 8, 124, 128]) - layer.4.v_cache: torch.Size([1, 8, 124, 128]) - layer.4.output: torch.Size([1, 124, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 124, 128]) - layer.0.v_cache: torch.Size([1, 8, 124, 128]) - layer.1.k_cache: torch.Size([1, 8, 124, 128]) - layer.1.v_cache: torch.Size([1, 8, 124, 128]) - layer.2.k_cache: torch.Size([1, 8, 124, 128]) - layer.2.v_cache: torch.Size([1, 8, 124, 128]) - layer.3.k_cache: torch.Size([1, 8, 124, 128]) - layer.3.v_cache: torch.Size([1, 8, 124, 128]) - layer.4.k_cache: torch.Size([1, 8, 124, 128]) - layer.4.v_cache: torch.Size([1, 8, 124, 128]) - layer.4.output: torch.Size([1, 124, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02574280 54.53665260 - layer.0.v_cache 0.00000027 0.00067160 - layer.1.k_cache 0.00324291 6.08873281 - layer.1.v_cache 0.00000080 0.00263280 - layer.2.k_cache 0.00117169 2.13127210 - layer.2.v_cache 0.00000108 0.00403920 - layer.3.k_cache 0.00137525 2.52412833 - layer.3.v_cache 0.00000203 0.00667880 - layer.4.k_cache 0.00340243 6.31697919 - layer.4.v_cache 0.00000299 0.01114985 - layer.4.output 0.00019780 0.23324064 - ------------------------------------------------------------------------------------- - TOTAL 0.00255239 5.18256427 - (elements=1,777,664) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1777664 -Total Bytes 15824 -BPFP 0.0712 bits/point -EBPFP 0.1424 equivalent bits/point -MSE 5.182564 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 124, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 124, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1826 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample38-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample38-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample388-layer4-item1.zst (53/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample388-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 85, 128) -Output shape: (1, 85, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.output: torch.Size([1, 85, 4096]) -> torch.Size([1, 1, 85, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 916B, BPFP=0.0842 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 928B, BPFP=0.0853 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,332B, BPFP=0.1224 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,100B, BPFP=0.1011 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,408B, BPFP=0.1294 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,232B, BPFP=0.1132 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,460B, BPFP=0.1342 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,280B, BPFP=0.1176 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,216B, BPFP=0.1118 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,212B, BPFP=0.1114 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,624B, BPFP=0.0833 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02690007 139.26530331 - layer.0.v_cache 0.00000028 0.00067307 - layer.1.k_cache 0.00337091 6.76798886 - layer.1.v_cache 0.00000080 0.00247222 - layer.2.k_cache 0.00114287 2.26353616 - layer.2.v_cache 0.00000105 0.00383599 - layer.3.k_cache 0.00139664 2.73507511 - layer.3.v_cache 0.00000196 0.00652942 - layer.4.k_cache 0.00332559 6.73472900 - layer.4.v_cache 0.00000294 0.01091817 - layer.4.output 0.00025832 0.26731150 - ------------------------------------------------------------------------------------- - TOTAL 0.00265546 11.34716481 - (elements=1,218,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1218560 -Total Bytes 15708 -BPFP 0.1031 bits/point -EBPFP 0.2062 equivalent bits/point -MSE 11.347165 ----------------------- -------------------------------------------------------- -Time: 0.488s Load: 0.005s, Pack+Encode: 0.203s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.3472 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample388-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample388-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample390-layer4-item1.zst (54/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample390-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 74, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 74, 128) -Output shape: (1, 74, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.output: torch.Size([1, 74, 4096]) -> torch.Size([1, 1, 74, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 972B, BPFP=0.1026 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,072B, BPFP=0.1132 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,324B, BPFP=0.1398 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,216B, BPFP=0.1284 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,412B, BPFP=0.1491 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,296B, BPFP=0.1368 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,436B, BPFP=0.1516 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,312B, BPFP=0.1385 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,232B, BPFP=0.1301 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,272B, BPFP=0.1343 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,976B, BPFP=0.1049 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.output: torch.Size([1, 74, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.output: torch.Size([1, 74, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02896725 106.51641681 - layer.0.v_cache 0.00000027 0.00068639 - layer.1.k_cache 0.00342263 7.13505925 - layer.1.v_cache 0.00000079 0.00283191 - layer.2.k_cache 0.00118243 2.34938544 - layer.2.v_cache 0.00000111 0.00417409 - layer.3.k_cache 0.00141981 2.81790532 - layer.3.v_cache 0.00000233 0.00713597 - layer.4.k_cache 0.00326747 6.79721234 - layer.4.v_cache 0.00000299 0.01119328 - layer.4.output 0.00023783 0.27377381 - ------------------------------------------------------------------------------------- - TOTAL 0.00280131 9.05264972 - (elements=1,060,864) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1060864 -Total Bytes 16520 -BPFP 0.1246 bits/point -EBPFP 0.2492 equivalent bits/point -MSE 9.052650 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.005s, Pack+Encode: 0.203s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 74, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.0526 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample390-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample390-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample40-layer4-item1.zst (55/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample40-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 126, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 126, 128) -Output shape: (1, 126, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.0.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.1.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.1.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.2.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.2.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.3.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.3.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.4.k_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.4.v_cache: torch.Size([1, 8, 126, 128]) -> torch.Size([1, 1, 126, 1024]) - layer.4.output: torch.Size([1, 126, 4096]) -> torch.Size([1, 1, 126, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 892B, BPFP=0.0553 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 892B, BPFP=0.0553 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,184B, BPFP=0.0734 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 988B, BPFP=0.0613 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,284B, BPFP=0.0796 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,072B, BPFP=0.0665 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,328B, BPFP=0.0823 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,112B, BPFP=0.0689 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,132B, BPFP=0.0702 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,076B, BPFP=0.0667 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,340B, BPFP=0.0518 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 126, 128]) - layer.0.v_cache: torch.Size([1, 8, 126, 128]) - layer.1.k_cache: torch.Size([1, 8, 126, 128]) - layer.1.v_cache: torch.Size([1, 8, 126, 128]) - layer.2.k_cache: torch.Size([1, 8, 126, 128]) - layer.2.v_cache: torch.Size([1, 8, 126, 128]) - layer.3.k_cache: torch.Size([1, 8, 126, 128]) - layer.3.v_cache: torch.Size([1, 8, 126, 128]) - layer.4.k_cache: torch.Size([1, 8, 126, 128]) - layer.4.v_cache: torch.Size([1, 8, 126, 128]) - layer.4.output: torch.Size([1, 126, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 126, 128]) - layer.0.v_cache: torch.Size([1, 8, 126, 128]) - layer.1.k_cache: torch.Size([1, 8, 126, 128]) - layer.1.v_cache: torch.Size([1, 8, 126, 128]) - layer.2.k_cache: torch.Size([1, 8, 126, 128]) - layer.2.v_cache: torch.Size([1, 8, 126, 128]) - layer.3.k_cache: torch.Size([1, 8, 126, 128]) - layer.3.v_cache: torch.Size([1, 8, 126, 128]) - layer.4.k_cache: torch.Size([1, 8, 126, 128]) - layer.4.v_cache: torch.Size([1, 8, 126, 128]) - layer.4.output: torch.Size([1, 126, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02675629 45.31772771 - layer.0.v_cache 0.00000026 0.00064011 - layer.1.k_cache 0.00313964 5.68055023 - layer.1.v_cache 0.00000077 0.00252314 - layer.2.k_cache 0.00115655 1.99332053 - layer.2.v_cache 0.00000107 0.00381748 - layer.3.k_cache 0.00144242 2.40139577 - layer.3.v_cache 0.00000205 0.00650452 - layer.4.k_cache 0.00340494 6.07539198 - layer.4.v_cache 0.00000295 0.01080717 - layer.4.output 0.00023638 0.22623448 - ------------------------------------------------------------------------------------- - TOTAL 0.00263232 4.45697261 - (elements=1,806,336) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1806336 -Total Bytes 14300 -BPFP 0.0633 bits/point -EBPFP 0.1267 equivalent bits/point -MSE 4.456973 ----------------------- -------------------------------------------------------- -Time: 0.495s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 126, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 126, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.4570 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample40-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample40-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample412-layer4-item1.zst (56/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample412-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 76, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 76, 128) -Output shape: (1, 76, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.0.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.1.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.1.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.2.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.2.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.3.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.3.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.4.k_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.4.v_cache: torch.Size([1, 8, 76, 128]) -> torch.Size([1, 1, 76, 1024]) - layer.4.output: torch.Size([1, 76, 4096]) -> torch.Size([1, 1, 76, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 960B, BPFP=0.0987 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,008B, BPFP=0.1036 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,320B, BPFP=0.1357 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,116B, BPFP=0.1147 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,380B, BPFP=0.1419 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,172B, BPFP=0.1205 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1480 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,284B, BPFP=0.1320 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,212B, BPFP=0.1246 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,212B, BPFP=0.1246 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,792B, BPFP=0.0975 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 76, 128]) - layer.0.v_cache: torch.Size([1, 8, 76, 128]) - layer.1.k_cache: torch.Size([1, 8, 76, 128]) - layer.1.v_cache: torch.Size([1, 8, 76, 128]) - layer.2.k_cache: torch.Size([1, 8, 76, 128]) - layer.2.v_cache: torch.Size([1, 8, 76, 128]) - layer.3.k_cache: torch.Size([1, 8, 76, 128]) - layer.3.v_cache: torch.Size([1, 8, 76, 128]) - layer.4.k_cache: torch.Size([1, 8, 76, 128]) - layer.4.v_cache: torch.Size([1, 8, 76, 128]) - layer.4.output: torch.Size([1, 76, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.278s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 76, 128]) - layer.0.v_cache: torch.Size([1, 8, 76, 128]) - layer.1.k_cache: torch.Size([1, 8, 76, 128]) - layer.1.v_cache: torch.Size([1, 8, 76, 128]) - layer.2.k_cache: torch.Size([1, 8, 76, 128]) - layer.2.v_cache: torch.Size([1, 8, 76, 128]) - layer.3.k_cache: torch.Size([1, 8, 76, 128]) - layer.3.v_cache: torch.Size([1, 8, 76, 128]) - layer.4.k_cache: torch.Size([1, 8, 76, 128]) - layer.4.v_cache: torch.Size([1, 8, 76, 128]) - layer.4.output: torch.Size([1, 76, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02706527 106.21666838 - layer.0.v_cache 0.00000027 0.00065571 - layer.1.k_cache 0.00360181 7.08791793 - layer.1.v_cache 0.00000073 0.00246108 - layer.2.k_cache 0.00113352 2.29726049 - layer.2.v_cache 0.00000108 0.00396585 - layer.3.k_cache 0.00139474 2.82903330 - layer.3.v_cache 0.00000224 0.00697638 - layer.4.k_cache 0.00318908 6.70933051 - layer.4.v_cache 0.00000296 0.01090807 - layer.4.output 0.00022408 0.29377732 - ------------------------------------------------------------------------------------- - TOTAL 0.00266343 9.02430621 - (elements=1,089,536) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1089536 -Total Bytes 15896 -BPFP 0.1167 bits/point -EBPFP 0.2334 equivalent bits/point -MSE 9.024306 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.005s, Pack+Encode: 0.204s, Decode+Unpack: 0.278s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 76, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 76, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.0243 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample412-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample412-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample414-layer4-item1.zst (57/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample414-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 85, 128) -Output shape: (1, 85, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) -> torch.Size([1, 1, 85, 1024]) - layer.4.output: torch.Size([1, 85, 4096]) -> torch.Size([1, 1, 85, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 920B, BPFP=0.0846 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,004B, BPFP=0.0923 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,340B, BPFP=0.1232 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,128B, BPFP=0.1037 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,420B, BPFP=0.1305 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,204B, BPFP=0.1107 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,472B, BPFP=0.1353 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,312B, BPFP=0.1206 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,224B, BPFP=0.1125 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,244B, BPFP=0.1143 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,572B, BPFP=0.0821 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 85, 128]) - layer.0.v_cache: torch.Size([1, 8, 85, 128]) - layer.1.k_cache: torch.Size([1, 8, 85, 128]) - layer.1.v_cache: torch.Size([1, 8, 85, 128]) - layer.2.k_cache: torch.Size([1, 8, 85, 128]) - layer.2.v_cache: torch.Size([1, 8, 85, 128]) - layer.3.k_cache: torch.Size([1, 8, 85, 128]) - layer.3.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.k_cache: torch.Size([1, 8, 85, 128]) - layer.4.v_cache: torch.Size([1, 8, 85, 128]) - layer.4.output: torch.Size([1, 85, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02461138 137.42635570 - layer.0.v_cache 0.00000028 0.00068982 - layer.1.k_cache 0.00340719 7.15675551 - layer.1.v_cache 0.00000073 0.00242048 - layer.2.k_cache 0.00115796 2.21676833 - layer.2.v_cache 0.00000109 0.00375652 - layer.3.k_cache 0.00135314 2.74447039 - layer.3.v_cache 0.00000196 0.00658575 - layer.4.k_cache 0.00328730 6.89921588 - layer.4.v_cache 0.00000286 0.01054582 - layer.4.output 0.00022769 0.25896353 - ------------------------------------------------------------------------------------- - TOTAL 0.00248105 11.25024417 - (elements=1,218,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1218560 -Total Bytes 15840 -BPFP 0.1040 bits/point -EBPFP 0.2080 equivalent bits/point -MSE 11.250244 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.007s, Pack+Encode: 0.203s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 85, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 85, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 11.2502 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample414-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample414-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample42-layer4-item1.zst (58/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample42-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 134, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 134, 128) -Output shape: (1, 134, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.0.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.1.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.1.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.2.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.2.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.3.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.3.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.4.k_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.4.v_cache: torch.Size([1, 8, 134, 128]) -> torch.Size([1, 1, 134, 1024]) - layer.4.output: torch.Size([1, 134, 4096]) -> torch.Size([1, 1, 134, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,604B, BPFP=0.0935 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,716B, BPFP=0.1000 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,440B, BPFP=0.1423 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,032B, BPFP=0.1185 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,492B, BPFP=0.1453 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,120B, BPFP=0.1236 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,468B, BPFP=0.1439 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,296B, BPFP=0.1339 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,208B, BPFP=0.1287 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,192B, BPFP=0.1278 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 6,100B, BPFP=0.0889 -⌛️ [2/4] FRONTEND: Frontend time: 0.248s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 134, 128]) - layer.0.v_cache: torch.Size([1, 8, 134, 128]) - layer.1.k_cache: torch.Size([1, 8, 134, 128]) - layer.1.v_cache: torch.Size([1, 8, 134, 128]) - layer.2.k_cache: torch.Size([1, 8, 134, 128]) - layer.2.v_cache: torch.Size([1, 8, 134, 128]) - layer.3.k_cache: torch.Size([1, 8, 134, 128]) - layer.3.v_cache: torch.Size([1, 8, 134, 128]) - layer.4.k_cache: torch.Size([1, 8, 134, 128]) - layer.4.v_cache: torch.Size([1, 8, 134, 128]) - layer.4.output: torch.Size([1, 134, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.372s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 134, 128]) - layer.0.v_cache: torch.Size([1, 8, 134, 128]) - layer.1.k_cache: torch.Size([1, 8, 134, 128]) - layer.1.v_cache: torch.Size([1, 8, 134, 128]) - layer.2.k_cache: torch.Size([1, 8, 134, 128]) - layer.2.v_cache: torch.Size([1, 8, 134, 128]) - layer.3.k_cache: torch.Size([1, 8, 134, 128]) - layer.3.v_cache: torch.Size([1, 8, 134, 128]) - layer.4.k_cache: torch.Size([1, 8, 134, 128]) - layer.4.v_cache: torch.Size([1, 8, 134, 128]) - layer.4.output: torch.Size([1, 134, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02623714 84.17839756 - layer.0.v_cache 0.00000029 0.00065506 - layer.1.k_cache 0.00318200 6.67978406 - layer.1.v_cache 0.00000078 0.00241726 - layer.2.k_cache 0.00115661 2.16301169 - layer.2.v_cache 0.00000102 0.00358997 - layer.3.k_cache 0.00142778 2.59377596 - layer.3.v_cache 0.00000209 0.00662331 - layer.4.k_cache 0.00344207 6.61322113 - layer.4.v_cache 0.00000292 0.01048852 - layer.4.output 0.00019652 0.22931701 - ------------------------------------------------------------------------------------- - TOTAL 0.00258848 7.36923090 - (elements=1,921,024) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1921024 -Total Bytes 27668 -BPFP 0.1152 bits/point -EBPFP 0.2304 equivalent bits/point -MSE 7.369231 ----------------------- -------------------------------------------------------- -Time: 0.628s Load: 0.007s, Pack+Encode: 0.248s, Decode+Unpack: 0.372s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 134, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 134, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.3692 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample42-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample42-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample44-layer4-item1.zst (59/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample44-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 132, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 132, 128) -Output shape: (1, 132, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.0.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.1.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.1.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.2.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.2.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.3.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.3.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.4.k_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.4.v_cache: torch.Size([1, 8, 132, 128]) -> torch.Size([1, 1, 132, 1024]) - layer.4.output: torch.Size([1, 132, 4096]) -> torch.Size([1, 1, 132, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,596B, BPFP=0.0945 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,684B, BPFP=0.0997 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,400B, BPFP=0.1420 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,984B, BPFP=0.1174 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,404B, BPFP=0.1423 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,160B, BPFP=0.1278 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,484B, BPFP=0.1470 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,176B, BPFP=0.1288 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,208B, BPFP=0.1307 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,132B, BPFP=0.1262 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 6,244B, BPFP=0.0924 -⌛️ [2/4] FRONTEND: Frontend time: 0.247s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 132, 128]) - layer.0.v_cache: torch.Size([1, 8, 132, 128]) - layer.1.k_cache: torch.Size([1, 8, 132, 128]) - layer.1.v_cache: torch.Size([1, 8, 132, 128]) - layer.2.k_cache: torch.Size([1, 8, 132, 128]) - layer.2.v_cache: torch.Size([1, 8, 132, 128]) - layer.3.k_cache: torch.Size([1, 8, 132, 128]) - layer.3.v_cache: torch.Size([1, 8, 132, 128]) - layer.4.k_cache: torch.Size([1, 8, 132, 128]) - layer.4.v_cache: torch.Size([1, 8, 132, 128]) - layer.4.output: torch.Size([1, 132, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.374s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 132, 128]) - layer.0.v_cache: torch.Size([1, 8, 132, 128]) - layer.1.k_cache: torch.Size([1, 8, 132, 128]) - layer.1.v_cache: torch.Size([1, 8, 132, 128]) - layer.2.k_cache: torch.Size([1, 8, 132, 128]) - layer.2.v_cache: torch.Size([1, 8, 132, 128]) - layer.3.k_cache: torch.Size([1, 8, 132, 128]) - layer.3.v_cache: torch.Size([1, 8, 132, 128]) - layer.4.k_cache: torch.Size([1, 8, 132, 128]) - layer.4.v_cache: torch.Size([1, 8, 132, 128]) - layer.4.output: torch.Size([1, 132, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02542288 74.22718395 - layer.0.v_cache 0.00000027 0.00065776 - layer.1.k_cache 0.00323066 6.50618813 - layer.1.v_cache 0.00000090 0.00263566 - layer.2.k_cache 0.00115984 2.14878105 - layer.2.v_cache 0.00000123 0.00422003 - layer.3.k_cache 0.00138390 2.62827648 - layer.3.v_cache 0.00000215 0.00727684 - layer.4.k_cache 0.00364779 6.61051987 - layer.4.v_cache 0.00000297 0.01126171 - layer.4.output 0.00020533 0.23774711 - ------------------------------------------------------------------------------------- - TOTAL 0.00254814 6.64985642 - (elements=1,892,352) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1892352 -Total Bytes 27472 -BPFP 0.1161 bits/point -EBPFP 0.2323 equivalent bits/point -MSE 6.649856 ----------------------- -------------------------------------------------------- -Time: 0.628s Load: 0.007s, Pack+Encode: 0.247s, Decode+Unpack: 0.374s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 132, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 132, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.6499 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample44-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample44-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample454-layer4-item1.zst (60/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample454-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 77, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 77, 128) -Output shape: (1, 77, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.output: torch.Size([1, 77, 4096]) -> torch.Size([1, 1, 77, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 940B, BPFP=0.0954 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,012B, BPFP=0.1027 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,332B, BPFP=0.1351 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,156B, BPFP=0.1173 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,384B, BPFP=0.1404 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,264B, BPFP=0.1282 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,436B, BPFP=0.1457 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,324B, BPFP=0.1343 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,220B, BPFP=0.1238 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,244B, BPFP=0.1262 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,824B, BPFP=0.0970 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.output: torch.Size([1, 77, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.output: torch.Size([1, 77, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02625101 112.05439580 - layer.0.v_cache 0.00000028 0.00069954 - layer.1.k_cache 0.00355533 7.21709175 - layer.1.v_cache 0.00000093 0.00280983 - layer.2.k_cache 0.00115642 2.32570262 - layer.2.v_cache 0.00000119 0.00447355 - layer.3.k_cache 0.00136722 2.87377058 - layer.3.v_cache 0.00000230 0.00760331 - layer.4.k_cache 0.00337541 6.90656310 - layer.4.v_cache 0.00000304 0.01184398 - layer.4.output 0.00023573 0.27113962 - ------------------------------------------------------------------------------------- - TOTAL 0.00261829 9.46353661 - (elements=1,103,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1103872 -Total Bytes 16136 -BPFP 0.1169 bits/point -EBPFP 0.2339 equivalent bits/point -MSE 9.463537 ----------------------- -------------------------------------------------------- -Time: 0.487s Load: 0.005s, Pack+Encode: 0.201s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 77, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.4635 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample454-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample454-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample464-layer4-item1.zst (61/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample464-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 94, 128) -Output shape: (1, 94, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) -> torch.Size([1, 1, 94, 1024]) - layer.4.output: torch.Size([1, 94, 4096]) -> torch.Size([1, 1, 94, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 992B, BPFP=0.0824 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,072B, BPFP=0.0891 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,396B, BPFP=0.1160 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,224B, BPFP=0.1017 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,456B, BPFP=0.1210 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,352B, BPFP=0.1124 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,560B, BPFP=0.1297 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,544B, BPFP=0.1283 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,380B, BPFP=0.1147 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,444B, BPFP=0.1200 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,208B, BPFP=0.0874 -⌛️ [2/4] FRONTEND: Frontend time: 0.200s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 94, 128]) - layer.0.v_cache: torch.Size([1, 8, 94, 128]) - layer.1.k_cache: torch.Size([1, 8, 94, 128]) - layer.1.v_cache: torch.Size([1, 8, 94, 128]) - layer.2.k_cache: torch.Size([1, 8, 94, 128]) - layer.2.v_cache: torch.Size([1, 8, 94, 128]) - layer.3.k_cache: torch.Size([1, 8, 94, 128]) - layer.3.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.k_cache: torch.Size([1, 8, 94, 128]) - layer.4.v_cache: torch.Size([1, 8, 94, 128]) - layer.4.output: torch.Size([1, 94, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02258782 80.57984957 - layer.0.v_cache 0.00000029 0.00067002 - layer.1.k_cache 0.00339276 6.87503376 - layer.1.v_cache 0.00000074 0.00230351 - layer.2.k_cache 0.00116230 2.11158493 - layer.2.v_cache 0.00000115 0.00374750 - layer.3.k_cache 0.00135573 2.60132518 - layer.3.v_cache 0.00000225 0.00701415 - layer.4.k_cache 0.00337629 6.60910813 - layer.4.v_cache 0.00000290 0.01051185 - layer.4.output 0.00017695 0.24441001 - ------------------------------------------------------------------------------------- - TOTAL 0.00232786 7.12705633 - (elements=1,347,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1347584 -Total Bytes 17628 -BPFP 0.1046 bits/point -EBPFP 0.2093 equivalent bits/point -MSE 7.127056 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.006s, Pack+Encode: 0.200s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 94, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 94, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.1271 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample464-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample464-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample478-layer4-item1.zst (62/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample478-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 77, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 77, 128) -Output shape: (1, 77, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) -> torch.Size([1, 1, 77, 1024]) - layer.4.output: torch.Size([1, 77, 4096]) -> torch.Size([1, 1, 77, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 944B, BPFP=0.0958 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 988B, BPFP=0.1002 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,308B, BPFP=0.1327 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,128B, BPFP=0.1144 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,392B, BPFP=0.1412 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,208B, BPFP=0.1226 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,420B, BPFP=0.1441 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,252B, BPFP=0.1270 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,204B, BPFP=0.1222 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,208B, BPFP=0.1226 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,700B, BPFP=0.0939 -⌛️ [2/4] FRONTEND: Frontend time: 0.200s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.output: torch.Size([1, 77, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 77, 128]) - layer.0.v_cache: torch.Size([1, 8, 77, 128]) - layer.1.k_cache: torch.Size([1, 8, 77, 128]) - layer.1.v_cache: torch.Size([1, 8, 77, 128]) - layer.2.k_cache: torch.Size([1, 8, 77, 128]) - layer.2.v_cache: torch.Size([1, 8, 77, 128]) - layer.3.k_cache: torch.Size([1, 8, 77, 128]) - layer.3.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.k_cache: torch.Size([1, 8, 77, 128]) - layer.4.v_cache: torch.Size([1, 8, 77, 128]) - layer.4.output: torch.Size([1, 77, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02707502 104.02206778 - layer.0.v_cache 0.00000027 0.00067591 - layer.1.k_cache 0.00339772 7.40829349 - layer.1.v_cache 0.00000076 0.00269807 - layer.2.k_cache 0.00114556 2.33501405 - layer.2.v_cache 0.00000112 0.00417841 - layer.3.k_cache 0.00146787 2.84248808 - layer.3.v_cache 0.00000204 0.00706085 - layer.4.k_cache 0.00318992 6.88665137 - layer.4.v_cache 0.00000300 0.01156758 - layer.4.output 0.00024419 0.29399057 - ------------------------------------------------------------------------------------- - TOTAL 0.00266143 8.90690413 - (elements=1,103,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1103872 -Total Bytes 15752 -BPFP 0.1142 bits/point -EBPFP 0.2283 equivalent bits/point -MSE 8.906904 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.005s, Pack+Encode: 0.200s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 77, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 77, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.9069 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample478-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample478-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample48-layer4-item1.zst (63/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample48-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 112, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 112, 128) -Output shape: (1, 112, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.output: torch.Size([1, 112, 4096]) -> torch.Size([1, 1, 112, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,100B, BPFP=0.0767 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,188B, BPFP=0.0829 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,552B, BPFP=0.1083 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,396B, BPFP=0.0974 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,568B, BPFP=0.1094 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,488B, BPFP=0.1038 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,604B, BPFP=0.1119 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,516B, BPFP=0.1057 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,472B, BPFP=0.1027 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,496B, BPFP=0.1044 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,808B, BPFP=0.0664 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.output: torch.Size([1, 112, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.output: torch.Size([1, 112, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02677278 93.86096191 - layer.0.v_cache 0.00000026 0.00064562 - layer.1.k_cache 0.00327242 6.61968449 - layer.1.v_cache 0.00000072 0.00247500 - layer.2.k_cache 0.00117792 2.24972616 - layer.2.v_cache 0.00000108 0.00390404 - layer.3.k_cache 0.00137547 2.69244548 - layer.3.v_cache 0.00000205 0.00668221 - layer.4.k_cache 0.00331726 6.78500530 - layer.4.v_cache 0.00000297 0.01087088 - layer.4.output 0.00024922 0.25061417 - ------------------------------------------------------------------------------------- - TOTAL 0.00263713 8.08820413 - (elements=1,605,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1605632 -Total Bytes 18188 -BPFP 0.0906 bits/point -EBPFP 0.1812 equivalent bits/point -MSE 8.088204 ----------------------- -------------------------------------------------------- -Time: 0.489s Load: 0.006s, Pack+Encode: 0.202s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 112, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.0882 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample48-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample48-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample485-layer4-item1.zst (64/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample485-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 78, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 78, 128) -Output shape: (1, 78, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.0.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.1.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.1.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.2.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.2.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.3.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.3.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.4.k_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.4.v_cache: torch.Size([1, 8, 78, 128]) -> torch.Size([1, 1, 78, 1024]) - layer.4.output: torch.Size([1, 78, 4096]) -> torch.Size([1, 1, 78, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 948B, BPFP=0.0950 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,004B, BPFP=0.1006 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,300B, BPFP=0.1302 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,144B, BPFP=0.1146 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,388B, BPFP=0.1390 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,216B, BPFP=0.1218 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,448B, BPFP=0.1450 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,296B, BPFP=0.1298 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,220B, BPFP=0.1222 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,228B, BPFP=0.1230 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,748B, BPFP=0.0939 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 78, 128]) - layer.0.v_cache: torch.Size([1, 8, 78, 128]) - layer.1.k_cache: torch.Size([1, 8, 78, 128]) - layer.1.v_cache: torch.Size([1, 8, 78, 128]) - layer.2.k_cache: torch.Size([1, 8, 78, 128]) - layer.2.v_cache: torch.Size([1, 8, 78, 128]) - layer.3.k_cache: torch.Size([1, 8, 78, 128]) - layer.3.v_cache: torch.Size([1, 8, 78, 128]) - layer.4.k_cache: torch.Size([1, 8, 78, 128]) - layer.4.v_cache: torch.Size([1, 8, 78, 128]) - layer.4.output: torch.Size([1, 78, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 78, 128]) - layer.0.v_cache: torch.Size([1, 8, 78, 128]) - layer.1.k_cache: torch.Size([1, 8, 78, 128]) - layer.1.v_cache: torch.Size([1, 8, 78, 128]) - layer.2.k_cache: torch.Size([1, 8, 78, 128]) - layer.2.v_cache: torch.Size([1, 8, 78, 128]) - layer.3.k_cache: torch.Size([1, 8, 78, 128]) - layer.3.v_cache: torch.Size([1, 8, 78, 128]) - layer.4.k_cache: torch.Size([1, 8, 78, 128]) - layer.4.v_cache: torch.Size([1, 8, 78, 128]) - layer.4.output: torch.Size([1, 78, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02543071 101.26774089 - layer.0.v_cache 0.00000030 0.00070202 - layer.1.k_cache 0.00350441 7.32372577 - layer.1.v_cache 0.00000082 0.00282343 - layer.2.k_cache 0.00114279 2.35365902 - layer.2.v_cache 0.00000132 0.00455411 - layer.3.k_cache 0.00135780 2.83884645 - layer.3.v_cache 0.00000231 0.00780913 - layer.4.k_cache 0.00328533 6.87298114 - layer.4.v_cache 0.00000329 0.01274432 - layer.4.output 0.00022272 0.26627550 - ------------------------------------------------------------------------------------- - TOTAL 0.00254428 8.69647774 - (elements=1,118,208) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1118208 -Total Bytes 15940 -BPFP 0.1140 bits/point -EBPFP 0.2281 equivalent bits/point -MSE 8.696478 ----------------------- -------------------------------------------------------- -Time: 0.484s Load: 0.005s, Pack+Encode: 0.201s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 78, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 78, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.6965 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample485-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample485-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample487-layer4-item1.zst (65/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample487-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 79, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 79, 128) -Output shape: (1, 79, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.0.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.1.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.1.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.2.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.2.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.3.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.3.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.4.k_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.4.v_cache: torch.Size([1, 8, 79, 128]) -> torch.Size([1, 1, 79, 1024]) - layer.4.output: torch.Size([1, 79, 4096]) -> torch.Size([1, 1, 79, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 924B, BPFP=0.0914 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 960B, BPFP=0.0949 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,304B, BPFP=0.1290 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,108B, BPFP=0.1096 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,392B, BPFP=0.1377 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,196B, BPFP=0.1183 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,432B, BPFP=0.1416 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,272B, BPFP=0.1258 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,216B, BPFP=0.1203 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,192B, BPFP=0.1179 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,716B, BPFP=0.0919 -⌛️ [2/4] FRONTEND: Frontend time: 0.200s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 79, 128]) - layer.0.v_cache: torch.Size([1, 8, 79, 128]) - layer.1.k_cache: torch.Size([1, 8, 79, 128]) - layer.1.v_cache: torch.Size([1, 8, 79, 128]) - layer.2.k_cache: torch.Size([1, 8, 79, 128]) - layer.2.v_cache: torch.Size([1, 8, 79, 128]) - layer.3.k_cache: torch.Size([1, 8, 79, 128]) - layer.3.v_cache: torch.Size([1, 8, 79, 128]) - layer.4.k_cache: torch.Size([1, 8, 79, 128]) - layer.4.v_cache: torch.Size([1, 8, 79, 128]) - layer.4.output: torch.Size([1, 79, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 79, 128]) - layer.0.v_cache: torch.Size([1, 8, 79, 128]) - layer.1.k_cache: torch.Size([1, 8, 79, 128]) - layer.1.v_cache: torch.Size([1, 8, 79, 128]) - layer.2.k_cache: torch.Size([1, 8, 79, 128]) - layer.2.v_cache: torch.Size([1, 8, 79, 128]) - layer.3.k_cache: torch.Size([1, 8, 79, 128]) - layer.3.v_cache: torch.Size([1, 8, 79, 128]) - layer.4.k_cache: torch.Size([1, 8, 79, 128]) - layer.4.v_cache: torch.Size([1, 8, 79, 128]) - layer.4.output: torch.Size([1, 79, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02536859 96.71475722 - layer.0.v_cache 0.00000029 0.00068381 - layer.1.k_cache 0.00331158 7.15575245 - layer.1.v_cache 0.00000077 0.00265502 - layer.2.k_cache 0.00112976 2.26681132 - layer.2.v_cache 0.00000111 0.00404573 - layer.3.k_cache 0.00133964 2.80745185 - layer.3.v_cache 0.00000223 0.00717724 - layer.4.k_cache 0.00332594 6.82271682 - layer.4.v_cache 0.00000299 0.01097611 - layer.4.output 0.00020400 0.24280430 - ------------------------------------------------------------------------------------- - TOTAL 0.00252135 8.34030320 - (elements=1,132,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1132544 -Total Bytes 15712 -BPFP 0.1110 bits/point -EBPFP 0.2220 equivalent bits/point -MSE 8.340303 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.005s, Pack+Encode: 0.200s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 79, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 79, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.3403 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample487-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample487-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample495-layer4-item1.zst (66/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample495-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 81, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 81, 128) -Output shape: (1, 81, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) -> torch.Size([1, 1, 81, 1024]) - layer.4.output: torch.Size([1, 81, 4096]) -> torch.Size([1, 1, 81, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 928B, BPFP=0.0895 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 972B, BPFP=0.0938 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,312B, BPFP=0.1265 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,136B, BPFP=0.1096 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,396B, BPFP=0.1346 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,220B, BPFP=0.1177 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1389 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,272B, BPFP=0.1227 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,188B, BPFP=0.1146 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,220B, BPFP=0.1177 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,784B, BPFP=0.0912 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.output: torch.Size([1, 81, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 81, 128]) - layer.0.v_cache: torch.Size([1, 8, 81, 128]) - layer.1.k_cache: torch.Size([1, 8, 81, 128]) - layer.1.v_cache: torch.Size([1, 8, 81, 128]) - layer.2.k_cache: torch.Size([1, 8, 81, 128]) - layer.2.v_cache: torch.Size([1, 8, 81, 128]) - layer.3.k_cache: torch.Size([1, 8, 81, 128]) - layer.3.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.k_cache: torch.Size([1, 8, 81, 128]) - layer.4.v_cache: torch.Size([1, 8, 81, 128]) - layer.4.output: torch.Size([1, 81, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02658518 86.12026789 - layer.0.v_cache 0.00000029 0.00069412 - layer.1.k_cache 0.00348817 7.07424211 - layer.1.v_cache 0.00000087 0.00276571 - layer.2.k_cache 0.00114702 2.28508655 - layer.2.v_cache 0.00000128 0.00454939 - layer.3.k_cache 0.00135614 2.76552591 - layer.3.v_cache 0.00000252 0.00770376 - layer.4.k_cache 0.00328397 6.57296188 - layer.4.v_cache 0.00000321 0.01201273 - layer.4.output 0.00022006 0.27175089 - ------------------------------------------------------------------------------------- - TOTAL 0.00262492 7.56662954 - (elements=1,161,216) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1161216 -Total Bytes 15868 -BPFP 0.1093 bits/point -EBPFP 0.2186 equivalent bits/point -MSE 7.566630 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.005s, Pack+Encode: 0.201s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 81, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 81, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.5666 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample495-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample495-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample50-layer4-item1.zst (67/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample50-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 117, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 117, 128) -Output shape: (1, 117, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.0.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.1.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.1.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.2.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.2.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.3.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.3.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.4.k_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.4.v_cache: torch.Size([1, 8, 117, 128]) -> torch.Size([1, 1, 117, 1024]) - layer.4.output: torch.Size([1, 117, 4096]) -> torch.Size([1, 1, 117, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,076B, BPFP=0.0718 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,168B, BPFP=0.0780 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,540B, BPFP=0.1028 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,312B, BPFP=0.0876 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,640B, BPFP=0.1095 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,460B, BPFP=0.0975 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,672B, BPFP=0.1116 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,568B, BPFP=0.1047 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,524B, BPFP=0.1018 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,504B, BPFP=0.1004 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,996B, BPFP=0.0667 -⌛️ [2/4] FRONTEND: Frontend time: 0.208s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 117, 128]) - layer.0.v_cache: torch.Size([1, 8, 117, 128]) - layer.1.k_cache: torch.Size([1, 8, 117, 128]) - layer.1.v_cache: torch.Size([1, 8, 117, 128]) - layer.2.k_cache: torch.Size([1, 8, 117, 128]) - layer.2.v_cache: torch.Size([1, 8, 117, 128]) - layer.3.k_cache: torch.Size([1, 8, 117, 128]) - layer.3.v_cache: torch.Size([1, 8, 117, 128]) - layer.4.k_cache: torch.Size([1, 8, 117, 128]) - layer.4.v_cache: torch.Size([1, 8, 117, 128]) - layer.4.output: torch.Size([1, 117, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 117, 128]) - layer.0.v_cache: torch.Size([1, 8, 117, 128]) - layer.1.k_cache: torch.Size([1, 8, 117, 128]) - layer.1.v_cache: torch.Size([1, 8, 117, 128]) - layer.2.k_cache: torch.Size([1, 8, 117, 128]) - layer.2.v_cache: torch.Size([1, 8, 117, 128]) - layer.3.k_cache: torch.Size([1, 8, 117, 128]) - layer.3.v_cache: torch.Size([1, 8, 117, 128]) - layer.4.k_cache: torch.Size([1, 8, 117, 128]) - layer.4.v_cache: torch.Size([1, 8, 117, 128]) - layer.4.output: torch.Size([1, 117, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02657393 119.93452858 - layer.0.v_cache 0.00000026 0.00066768 - layer.1.k_cache 0.00326448 6.90552071 - layer.1.v_cache 0.00000079 0.00273493 - layer.2.k_cache 0.00117087 2.26197854 - layer.2.v_cache 0.00000122 0.00432491 - layer.3.k_cache 0.00141366 2.70915679 - layer.3.v_cache 0.00000241 0.00742260 - layer.4.k_cache 0.00339367 6.90710345 - layer.4.v_cache 0.00000311 0.01176865 - layer.4.output 0.00025967 0.26939392 - ------------------------------------------------------------------------------------- - TOTAL 0.00263308 9.98734161 - (elements=1,677,312) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1677312 -Total Bytes 18460 -BPFP 0.0880 bits/point -EBPFP 0.1761 equivalent bits/point -MSE 9.987342 ----------------------- -------------------------------------------------------- -Time: 0.498s Load: 0.006s, Pack+Encode: 0.208s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 117, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 117, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.9873 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample50-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample50-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample516-layer4-item1.zst (68/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample516-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 84, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 84, 128) -Output shape: (1, 84, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.0.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.1.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.1.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.2.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.2.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.3.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.3.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.4.k_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.4.v_cache: torch.Size([1, 8, 84, 128]) -> torch.Size([1, 1, 84, 1024]) - layer.4.output: torch.Size([1, 84, 4096]) -> torch.Size([1, 1, 84, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 920B, BPFP=0.0856 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 960B, BPFP=0.0893 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,328B, BPFP=0.1235 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,132B, BPFP=0.1053 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,412B, BPFP=0.1313 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,188B, BPFP=0.1105 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,460B, BPFP=0.1358 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,288B, BPFP=0.1198 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,216B, BPFP=0.1131 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,260B, BPFP=0.1172 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,868B, BPFP=0.0899 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 84, 128]) - layer.0.v_cache: torch.Size([1, 8, 84, 128]) - layer.1.k_cache: torch.Size([1, 8, 84, 128]) - layer.1.v_cache: torch.Size([1, 8, 84, 128]) - layer.2.k_cache: torch.Size([1, 8, 84, 128]) - layer.2.v_cache: torch.Size([1, 8, 84, 128]) - layer.3.k_cache: torch.Size([1, 8, 84, 128]) - layer.3.v_cache: torch.Size([1, 8, 84, 128]) - layer.4.k_cache: torch.Size([1, 8, 84, 128]) - layer.4.v_cache: torch.Size([1, 8, 84, 128]) - layer.4.output: torch.Size([1, 84, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 84, 128]) - layer.0.v_cache: torch.Size([1, 8, 84, 128]) - layer.1.k_cache: torch.Size([1, 8, 84, 128]) - layer.1.v_cache: torch.Size([1, 8, 84, 128]) - layer.2.k_cache: torch.Size([1, 8, 84, 128]) - layer.2.v_cache: torch.Size([1, 8, 84, 128]) - layer.3.k_cache: torch.Size([1, 8, 84, 128]) - layer.3.v_cache: torch.Size([1, 8, 84, 128]) - layer.4.k_cache: torch.Size([1, 8, 84, 128]) - layer.4.v_cache: torch.Size([1, 8, 84, 128]) - layer.4.output: torch.Size([1, 84, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02585370 116.18065244 - layer.0.v_cache 0.00000028 0.00066252 - layer.1.k_cache 0.00342466 7.02605838 - layer.1.v_cache 0.00000080 0.00270306 - layer.2.k_cache 0.00116354 2.25922957 - layer.2.v_cache 0.00000112 0.00407361 - layer.3.k_cache 0.00133515 2.72165462 - layer.3.v_cache 0.00000214 0.00719056 - layer.4.k_cache 0.00358854 6.95672026 - layer.4.v_cache 0.00000313 0.01165366 - layer.4.output 0.00016584 0.23952184 - ------------------------------------------------------------------------------------- - TOTAL 0.00257403 9.72347758 - (elements=1,204,224) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1204224 -Total Bytes 16032 -BPFP 0.1065 bits/point -EBPFP 0.2130 equivalent bits/point -MSE 9.723478 ----------------------- -------------------------------------------------------- -Time: 0.487s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 84, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 84, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.7235 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample516-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample516-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample52-layer4-item1.zst (69/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample52-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 133, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 133, 128) -Output shape: (1, 133, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.0.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.1.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.1.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.2.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.2.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.3.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.3.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.4.k_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.4.v_cache: torch.Size([1, 8, 133, 128]) -> torch.Size([1, 1, 133, 1024]) - layer.4.output: torch.Size([1, 133, 4096]) -> torch.Size([1, 1, 133, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,584B, BPFP=0.0930 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,752B, BPFP=0.1029 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,352B, BPFP=0.1382 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,028B, BPFP=0.1191 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,440B, BPFP=0.1433 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,168B, BPFP=0.1273 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,496B, BPFP=0.1466 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,228B, BPFP=0.1309 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,264B, BPFP=0.1330 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,160B, BPFP=0.1269 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 6,384B, BPFP=0.0938 -⌛️ [2/4] FRONTEND: Frontend time: 0.249s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 133, 128]) - layer.0.v_cache: torch.Size([1, 8, 133, 128]) - layer.1.k_cache: torch.Size([1, 8, 133, 128]) - layer.1.v_cache: torch.Size([1, 8, 133, 128]) - layer.2.k_cache: torch.Size([1, 8, 133, 128]) - layer.2.v_cache: torch.Size([1, 8, 133, 128]) - layer.3.k_cache: torch.Size([1, 8, 133, 128]) - layer.3.v_cache: torch.Size([1, 8, 133, 128]) - layer.4.k_cache: torch.Size([1, 8, 133, 128]) - layer.4.v_cache: torch.Size([1, 8, 133, 128]) - layer.4.output: torch.Size([1, 133, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.377s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 133, 128]) - layer.0.v_cache: torch.Size([1, 8, 133, 128]) - layer.1.k_cache: torch.Size([1, 8, 133, 128]) - layer.1.v_cache: torch.Size([1, 8, 133, 128]) - layer.2.k_cache: torch.Size([1, 8, 133, 128]) - layer.2.v_cache: torch.Size([1, 8, 133, 128]) - layer.3.k_cache: torch.Size([1, 8, 133, 128]) - layer.3.v_cache: torch.Size([1, 8, 133, 128]) - layer.4.k_cache: torch.Size([1, 8, 133, 128]) - layer.4.v_cache: torch.Size([1, 8, 133, 128]) - layer.4.output: torch.Size([1, 133, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02254625 83.52466371 - layer.0.v_cache 0.00000028 0.00068715 - layer.1.k_cache 0.00324298 6.30276788 - layer.1.v_cache 0.00000087 0.00256588 - layer.2.k_cache 0.00114197 2.14270800 - layer.2.v_cache 0.00000110 0.00410616 - layer.3.k_cache 0.00134409 2.63196443 - layer.3.v_cache 0.00000211 0.00705066 - layer.4.k_cache 0.00375619 6.83402384 - layer.4.v_cache 0.00000306 0.01134396 - layer.4.output 0.00014515 0.21475785 - ------------------------------------------------------------------------------------- - TOTAL 0.00232996 7.30863665 - (elements=1,906,688) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1906688 -Total Bytes 27856 -BPFP 0.1169 bits/point -EBPFP 0.2338 equivalent bits/point -MSE 7.308637 ----------------------- -------------------------------------------------------- -Time: 0.633s Load: 0.008s, Pack+Encode: 0.249s, Decode+Unpack: 0.377s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 133, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 133, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.3086 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample52-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample52-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample53-layer4-item1.zst (70/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample53-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 125, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 125, 128) -Output shape: (1, 125, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.0.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.1.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.1.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.2.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.2.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.3.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.3.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.4.k_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.4.v_cache: torch.Size([1, 8, 125, 128]) -> torch.Size([1, 1, 125, 1024]) - layer.4.output: torch.Size([1, 125, 4096]) -> torch.Size([1, 1, 125, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 980B, BPFP=0.0612 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 984B, BPFP=0.0615 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,236B, BPFP=0.0772 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,076B, BPFP=0.0673 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,332B, BPFP=0.0833 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,140B, BPFP=0.0712 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,376B, BPFP=0.0860 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,196B, BPFP=0.0747 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,196B, BPFP=0.0747 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,152B, BPFP=0.0720 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,488B, BPFP=0.0545 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 125, 128]) - layer.0.v_cache: torch.Size([1, 8, 125, 128]) - layer.1.k_cache: torch.Size([1, 8, 125, 128]) - layer.1.v_cache: torch.Size([1, 8, 125, 128]) - layer.2.k_cache: torch.Size([1, 8, 125, 128]) - layer.2.v_cache: torch.Size([1, 8, 125, 128]) - layer.3.k_cache: torch.Size([1, 8, 125, 128]) - layer.3.v_cache: torch.Size([1, 8, 125, 128]) - layer.4.k_cache: torch.Size([1, 8, 125, 128]) - layer.4.v_cache: torch.Size([1, 8, 125, 128]) - layer.4.output: torch.Size([1, 125, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 125, 128]) - layer.0.v_cache: torch.Size([1, 8, 125, 128]) - layer.1.k_cache: torch.Size([1, 8, 125, 128]) - layer.1.v_cache: torch.Size([1, 8, 125, 128]) - layer.2.k_cache: torch.Size([1, 8, 125, 128]) - layer.2.v_cache: torch.Size([1, 8, 125, 128]) - layer.3.k_cache: torch.Size([1, 8, 125, 128]) - layer.3.v_cache: torch.Size([1, 8, 125, 128]) - layer.4.k_cache: torch.Size([1, 8, 125, 128]) - layer.4.v_cache: torch.Size([1, 8, 125, 128]) - layer.4.output: torch.Size([1, 125, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02456201 47.88385938 - layer.0.v_cache 0.00000027 0.00063140 - layer.1.k_cache 0.00318185 6.06568896 - layer.1.v_cache 0.00000072 0.00244960 - layer.2.k_cache 0.00113729 2.03661157 - layer.2.v_cache 0.00000107 0.00389814 - layer.3.k_cache 0.00136705 2.46837402 - layer.3.v_cache 0.00000203 0.00671149 - layer.4.k_cache 0.00364986 6.32924707 - layer.4.v_cache 0.00000297 0.01054428 - layer.4.output 0.00019454 0.22140892 - ------------------------------------------------------------------------------------- - TOTAL 0.00247738 4.69240369 - (elements=1,792,000) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1792000 -Total Bytes 15156 -BPFP 0.0677 bits/point -EBPFP 0.1353 equivalent bits/point -MSE 4.692404 ----------------------- -------------------------------------------------------- -Time: 0.494s Load: 0.008s, Pack+Encode: 0.203s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 125, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 125, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.6924 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample53-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample53-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample543-layer4-item1.zst (71/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample543-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 71, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 71, 128) -Output shape: (1, 71, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.output: torch.Size([1, 71, 4096]) -> torch.Size([1, 1, 71, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,016B, BPFP=0.1118 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,056B, BPFP=0.1162 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,356B, BPFP=0.1492 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,224B, BPFP=0.1347 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,420B, BPFP=0.1562 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,288B, BPFP=0.1417 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1585 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,320B, BPFP=0.1452 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,268B, BPFP=0.1395 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,288B, BPFP=0.1417 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,316B, BPFP=0.1187 -⌛️ [2/4] FRONTEND: Frontend time: 0.205s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.output: torch.Size([1, 71, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.output: torch.Size([1, 71, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02849298 120.59800011 - layer.0.v_cache 0.00000028 0.00066422 - layer.1.k_cache 0.00376352 7.49927703 - layer.1.v_cache 0.00000080 0.00276391 - layer.2.k_cache 0.00117123 2.31342154 - layer.2.v_cache 0.00000118 0.00415186 - layer.3.k_cache 0.00138233 2.82931218 - layer.3.v_cache 0.00000223 0.00732169 - layer.4.k_cache 0.00337686 7.00823545 - layer.4.v_cache 0.00000300 0.01158428 - layer.4.output 0.00019204 0.25825629 - ------------------------------------------------------------------------------------- - TOTAL 0.00278304 10.09341125 - (elements=1,017,856) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1017856 -Total Bytes 16992 -BPFP 0.1336 bits/point -EBPFP 0.2671 equivalent bits/point -MSE 10.093411 ----------------------- -------------------------------------------------------- -Time: 0.491s Load: 0.004s, Pack+Encode: 0.205s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 71, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.0934 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample543-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample543-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample561-layer4-item1.zst (72/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample561-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 68, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 68, 128) -Output shape: (1, 68, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.output: torch.Size([1, 68, 4096]) -> torch.Size([1, 1, 68, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,024B, BPFP=0.1176 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,056B, BPFP=0.1213 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,360B, BPFP=0.1562 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,232B, BPFP=0.1415 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,420B, BPFP=0.1631 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,284B, BPFP=0.1475 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1654 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,316B, BPFP=0.1512 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,268B, BPFP=0.1457 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,272B, BPFP=0.1461 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,912B, BPFP=0.1124 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.output: torch.Size([1, 68, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.output: torch.Size([1, 68, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02799314 109.60938936 - layer.0.v_cache 0.00000027 0.00069045 - layer.1.k_cache 0.00364027 7.42242925 - layer.1.v_cache 0.00000080 0.00295947 - layer.2.k_cache 0.00115812 2.41419467 - layer.2.v_cache 0.00000117 0.00453295 - layer.3.k_cache 0.00137295 2.85982042 - layer.3.v_cache 0.00000227 0.00803611 - layer.4.k_cache 0.00327663 6.82971730 - layer.4.v_cache 0.00000341 0.01287012 - layer.4.output 0.00018782 0.26324976 - ------------------------------------------------------------------------------------- - TOTAL 0.00272859 9.30125994 - (elements=974,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 974848 -Total Bytes 16584 -BPFP 0.1361 bits/point -EBPFP 0.2722 equivalent bits/point -MSE 9.301260 ----------------------- -------------------------------------------------------- -Time: 0.487s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 68, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.3013 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample561-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample561-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample570-layer4-item1.zst (73/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample570-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 72, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 72, 128) -Output shape: (1, 72, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.output: torch.Size([1, 72, 4096]) -> torch.Size([1, 1, 72, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,036B, BPFP=0.1124 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,100B, BPFP=0.1194 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,364B, BPFP=0.1480 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,212B, BPFP=0.1315 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,412B, BPFP=0.1532 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,268B, BPFP=0.1376 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,428B, BPFP=0.1549 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,324B, BPFP=0.1437 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,256B, BPFP=0.1363 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,240B, BPFP=0.1345 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,280B, BPFP=0.1161 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.output: torch.Size([1, 72, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.output: torch.Size([1, 72, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02695644 96.61518690 - layer.0.v_cache 0.00000028 0.00067694 - layer.1.k_cache 0.00362044 7.23887380 - layer.1.v_cache 0.00000088 0.00268696 - layer.2.k_cache 0.00116887 2.31834136 - layer.2.v_cache 0.00000122 0.00424778 - layer.3.k_cache 0.00138034 2.78663699 - layer.3.v_cache 0.00000238 0.00731922 - layer.4.k_cache 0.00335768 6.84859085 - layer.4.v_cache 0.00000294 0.01083444 - layer.4.output 0.00021991 0.25330745 - ------------------------------------------------------------------------------------- - TOTAL 0.00266936 8.34618750 - (elements=1,032,192) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1032192 -Total Bytes 16920 -BPFP 0.1311 bits/point -EBPFP 0.2623 equivalent bits/point -MSE 8.346188 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.005s, Pack+Encode: 0.203s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 72, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.3462 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample570-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample570-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample581-layer4-item1.zst (74/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample581-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 73, 128) -Output shape: (1, 73, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.output: torch.Size([1, 73, 4096]) -> torch.Size([1, 1, 73, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,016B, BPFP=0.1087 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,044B, BPFP=0.1117 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,348B, BPFP=0.1443 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,160B, BPFP=0.1241 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,408B, BPFP=0.1507 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,248B, BPFP=0.1336 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,444B, BPFP=0.1545 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,316B, BPFP=0.1408 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,252B, BPFP=0.1340 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,256B, BPFP=0.1344 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,092B, BPFP=0.1095 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.280s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02610407 93.20432898 - layer.0.v_cache 0.00000028 0.00068613 - layer.1.k_cache 0.00357364 7.14237077 - layer.1.v_cache 0.00000075 0.00263753 - layer.2.k_cache 0.00114681 2.30323687 - layer.2.v_cache 0.00000108 0.00398263 - layer.3.k_cache 0.00136272 2.78696682 - layer.3.v_cache 0.00000212 0.00685059 - layer.4.k_cache 0.00341161 6.74733870 - layer.4.v_cache 0.00000293 0.01113332 - layer.4.output 0.00018354 0.25855718 - ------------------------------------------------------------------------------------- - TOTAL 0.00259573 8.08884007 - (elements=1,046,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1046528 -Total Bytes 16584 -BPFP 0.1268 bits/point -EBPFP 0.2535 equivalent bits/point -MSE 8.088840 ----------------------- -------------------------------------------------------- -Time: 0.486s Load: 0.005s, Pack+Encode: 0.201s, Decode+Unpack: 0.280s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.0888 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample581-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample581-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample584-layer4-item1.zst (75/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample584-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 72, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 72, 128) -Output shape: (1, 72, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) -> torch.Size([1, 1, 72, 1024]) - layer.4.output: torch.Size([1, 72, 4096]) -> torch.Size([1, 1, 72, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,008B, BPFP=0.1094 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,076B, BPFP=0.1168 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,344B, BPFP=0.1458 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,196B, BPFP=0.1298 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,408B, BPFP=0.1528 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,256B, BPFP=0.1363 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,444B, BPFP=0.1567 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,316B, BPFP=0.1428 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,256B, BPFP=0.1363 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,272B, BPFP=0.1380 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,204B, BPFP=0.1140 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.output: torch.Size([1, 72, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 72, 128]) - layer.0.v_cache: torch.Size([1, 8, 72, 128]) - layer.1.k_cache: torch.Size([1, 8, 72, 128]) - layer.1.v_cache: torch.Size([1, 8, 72, 128]) - layer.2.k_cache: torch.Size([1, 8, 72, 128]) - layer.2.v_cache: torch.Size([1, 8, 72, 128]) - layer.3.k_cache: torch.Size([1, 8, 72, 128]) - layer.3.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.k_cache: torch.Size([1, 8, 72, 128]) - layer.4.v_cache: torch.Size([1, 8, 72, 128]) - layer.4.output: torch.Size([1, 72, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02727311 105.23109945 - layer.0.v_cache 0.00000029 0.00067691 - layer.1.k_cache 0.00335621 7.18180423 - layer.1.v_cache 0.00000072 0.00244079 - layer.2.k_cache 0.00111533 2.25485335 - layer.2.v_cache 0.00000112 0.00394912 - layer.3.k_cache 0.00138591 2.74593353 - layer.3.v_cache 0.00000206 0.00675971 - layer.4.k_cache 0.00323834 6.73134274 - layer.4.v_cache 0.00000293 0.01082387 - layer.4.output 0.00022510 0.24434198 - ------------------------------------------------------------------------------------- - TOTAL 0.00266260 8.93907512 - (elements=1,032,192) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1032192 -Total Bytes 16780 -BPFP 0.1301 bits/point -EBPFP 0.2601 equivalent bits/point -MSE 8.939075 ----------------------- -------------------------------------------------------- -Time: 0.487s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 72, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 72, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.9391 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample584-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample584-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample60-layer4-item1.zst (76/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample60-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 115, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 115, 128) -Output shape: (1, 115, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.0.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.1.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.1.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.2.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.2.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.3.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.3.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.4.k_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.4.v_cache: torch.Size([1, 8, 115, 128]) -> torch.Size([1, 1, 115, 1024]) - layer.4.output: torch.Size([1, 115, 4096]) -> torch.Size([1, 1, 115, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,096B, BPFP=0.0745 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,120B, BPFP=0.0761 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,552B, BPFP=0.1054 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,412B, BPFP=0.0959 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,624B, BPFP=0.1103 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,452B, BPFP=0.0986 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,652B, BPFP=0.1122 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,504B, BPFP=0.1022 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,496B, BPFP=0.1016 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,476B, BPFP=0.1003 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,960B, BPFP=0.0673 -⌛️ [2/4] FRONTEND: Frontend time: 0.203s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 115, 128]) - layer.0.v_cache: torch.Size([1, 8, 115, 128]) - layer.1.k_cache: torch.Size([1, 8, 115, 128]) - layer.1.v_cache: torch.Size([1, 8, 115, 128]) - layer.2.k_cache: torch.Size([1, 8, 115, 128]) - layer.2.v_cache: torch.Size([1, 8, 115, 128]) - layer.3.k_cache: torch.Size([1, 8, 115, 128]) - layer.3.v_cache: torch.Size([1, 8, 115, 128]) - layer.4.k_cache: torch.Size([1, 8, 115, 128]) - layer.4.v_cache: torch.Size([1, 8, 115, 128]) - layer.4.output: torch.Size([1, 115, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 115, 128]) - layer.0.v_cache: torch.Size([1, 8, 115, 128]) - layer.1.k_cache: torch.Size([1, 8, 115, 128]) - layer.1.v_cache: torch.Size([1, 8, 115, 128]) - layer.2.k_cache: torch.Size([1, 8, 115, 128]) - layer.2.v_cache: torch.Size([1, 8, 115, 128]) - layer.3.k_cache: torch.Size([1, 8, 115, 128]) - layer.3.v_cache: torch.Size([1, 8, 115, 128]) - layer.4.k_cache: torch.Size([1, 8, 115, 128]) - layer.4.v_cache: torch.Size([1, 8, 115, 128]) - layer.4.output: torch.Size([1, 115, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02486876 123.27300442 - layer.0.v_cache 0.00000027 0.00068433 - layer.1.k_cache 0.00345256 6.75872644 - layer.1.v_cache 0.00000081 0.00278316 - layer.2.k_cache 0.00114556 2.23281860 - layer.2.v_cache 0.00000111 0.00421402 - layer.3.k_cache 0.00139466 2.70503673 - layer.3.v_cache 0.00000215 0.00729422 - layer.4.k_cache 0.00339940 6.68532927 - layer.4.v_cache 0.00000318 0.01201785 - layer.4.output 0.00018696 0.23736995 - ------------------------------------------------------------------------------------- - TOTAL 0.00250116 10.18795635 - (elements=1,648,640) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1648640 -Total Bytes 18344 -BPFP 0.0890 bits/point -EBPFP 0.1780 equivalent bits/point -MSE 10.187956 ----------------------- -------------------------------------------------------- -Time: 0.492s Load: 0.007s, Pack+Encode: 0.203s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 115, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 115, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.1880 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample60-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample60-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample600-layer4-item1.zst (77/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample600-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 71, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 71, 128) -Output shape: (1, 71, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) -> torch.Size([1, 1, 71, 1024]) - layer.4.output: torch.Size([1, 71, 4096]) -> torch.Size([1, 1, 71, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,016B, BPFP=0.1118 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,056B, BPFP=0.1162 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,356B, BPFP=0.1492 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,188B, BPFP=0.1307 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,392B, BPFP=0.1532 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,248B, BPFP=0.1373 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,424B, BPFP=0.1567 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,324B, BPFP=0.1457 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,244B, BPFP=0.1369 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,248B, BPFP=0.1373 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,052B, BPFP=0.1115 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.output: torch.Size([1, 71, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 71, 128]) - layer.0.v_cache: torch.Size([1, 8, 71, 128]) - layer.1.k_cache: torch.Size([1, 8, 71, 128]) - layer.1.v_cache: torch.Size([1, 8, 71, 128]) - layer.2.k_cache: torch.Size([1, 8, 71, 128]) - layer.2.v_cache: torch.Size([1, 8, 71, 128]) - layer.3.k_cache: torch.Size([1, 8, 71, 128]) - layer.3.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.k_cache: torch.Size([1, 8, 71, 128]) - layer.4.v_cache: torch.Size([1, 8, 71, 128]) - layer.4.output: torch.Size([1, 71, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02663550 121.46471996 - layer.0.v_cache 0.00000027 0.00068827 - layer.1.k_cache 0.00355294 7.28925076 - layer.1.v_cache 0.00000089 0.00279823 - layer.2.k_cache 0.00117088 2.32212808 - layer.2.v_cache 0.00000116 0.00429685 - layer.3.k_cache 0.00139091 2.85488268 - layer.3.v_cache 0.00000218 0.00759763 - layer.4.k_cache 0.00325425 6.82143563 - layer.4.v_cache 0.00000306 0.01153368 - layer.4.output 0.00022008 0.26776470 - ------------------------------------------------------------------------------------- - TOTAL 0.00263517 10.13217075 - (elements=1,017,856) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1017856 -Total Bytes 16548 -BPFP 0.1301 bits/point -EBPFP 0.2601 equivalent bits/point -MSE 10.132171 ----------------------- -------------------------------------------------------- -Time: 0.487s Load: 0.004s, Pack+Encode: 0.201s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 71, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 71, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.1322 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample600-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample600-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample622-layer4-item1.zst (78/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample622-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 73, 128) -Output shape: (1, 73, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.output: torch.Size([1, 73, 4096]) -> torch.Size([1, 1, 73, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,000B, BPFP=0.1070 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,072B, BPFP=0.1147 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,348B, BPFP=0.1443 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,236B, BPFP=0.1323 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,392B, BPFP=0.1490 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,304B, BPFP=0.1396 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1541 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,328B, BPFP=0.1421 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,228B, BPFP=0.1314 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,292B, BPFP=0.1383 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,068B, BPFP=0.1088 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02450750 99.25421393 - layer.0.v_cache 0.00000028 0.00067839 - layer.1.k_cache 0.00347943 7.00665032 - layer.1.v_cache 0.00000079 0.00272526 - layer.2.k_cache 0.00114165 2.28285614 - layer.2.v_cache 0.00000118 0.00430633 - layer.3.k_cache 0.00136377 2.80280523 - layer.3.v_cache 0.00000212 0.00730533 - layer.4.k_cache 0.00321585 6.56349078 - layer.4.v_cache 0.00000299 0.01142284 - layer.4.output 0.00023923 0.27189963 - ------------------------------------------------------------------------------------- - TOTAL 0.00247661 8.50171808 - (elements=1,046,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1046528 -Total Bytes 16708 -BPFP 0.1277 bits/point -EBPFP 0.2554 equivalent bits/point -MSE 8.501718 ----------------------- -------------------------------------------------------- -Time: 0.490s Load: 0.005s, Pack+Encode: 0.202s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.5017 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample622-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample622-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample65-layer4-item1.zst (79/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample65-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 112, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 112, 128) -Output shape: (1, 112, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) -> torch.Size([1, 1, 112, 1024]) - layer.4.output: torch.Size([1, 112, 4096]) -> torch.Size([1, 1, 112, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,084B, BPFP=0.0756 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,164B, BPFP=0.0812 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,520B, BPFP=0.1060 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,320B, BPFP=0.0921 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,540B, BPFP=0.1074 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,488B, BPFP=0.1038 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,616B, BPFP=0.1127 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,480B, BPFP=0.1032 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,416B, BPFP=0.0988 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,424B, BPFP=0.0993 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,284B, BPFP=0.0747 -⌛️ [2/4] FRONTEND: Frontend time: 0.202s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.output: torch.Size([1, 112, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.271s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 112, 128]) - layer.0.v_cache: torch.Size([1, 8, 112, 128]) - layer.1.k_cache: torch.Size([1, 8, 112, 128]) - layer.1.v_cache: torch.Size([1, 8, 112, 128]) - layer.2.k_cache: torch.Size([1, 8, 112, 128]) - layer.2.v_cache: torch.Size([1, 8, 112, 128]) - layer.3.k_cache: torch.Size([1, 8, 112, 128]) - layer.3.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.k_cache: torch.Size([1, 8, 112, 128]) - layer.4.v_cache: torch.Size([1, 8, 112, 128]) - layer.4.output: torch.Size([1, 112, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02850304 100.86040388 - layer.0.v_cache 0.00000027 0.00069914 - layer.1.k_cache 0.00329348 6.57974298 - layer.1.v_cache 0.00000088 0.00300779 - layer.2.k_cache 0.00114370 2.21131352 - layer.2.v_cache 0.00000135 0.00441471 - layer.3.k_cache 0.00133332 2.72352164 - layer.3.v_cache 0.00000257 0.00794609 - layer.4.k_cache 0.00339413 6.59961482 - layer.4.v_cache 0.00000321 0.01210010 - layer.4.output 0.00019581 0.24123883 - ------------------------------------------------------------------------------------- - TOTAL 0.00274709 8.56912286 - (elements=1,605,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1605632 -Total Bytes 18336 -BPFP 0.0914 bits/point -EBPFP 0.1827 equivalent bits/point -MSE 8.569123 ----------------------- -------------------------------------------------------- -Time: 0.480s Load: 0.007s, Pack+Encode: 0.202s, Decode+Unpack: 0.271s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 112, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 112, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.5691 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample65-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample65-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample656-layer4-item1.zst (80/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample656-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 68, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 68, 128) -Output shape: (1, 68, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) -> torch.Size([1, 1, 68, 1024]) - layer.4.output: torch.Size([1, 68, 4096]) -> torch.Size([1, 1, 68, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,020B, BPFP=0.1172 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,068B, BPFP=0.1227 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,352B, BPFP=0.1553 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,236B, BPFP=0.1420 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,416B, BPFP=0.1627 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,300B, BPFP=0.1494 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,444B, BPFP=0.1659 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,332B, BPFP=0.1530 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,288B, BPFP=0.1480 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,288B, BPFP=0.1480 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,044B, BPFP=0.1162 -⌛️ [2/4] FRONTEND: Frontend time: 0.195s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.output: torch.Size([1, 68, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.271s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 68, 128]) - layer.0.v_cache: torch.Size([1, 8, 68, 128]) - layer.1.k_cache: torch.Size([1, 8, 68, 128]) - layer.1.v_cache: torch.Size([1, 8, 68, 128]) - layer.2.k_cache: torch.Size([1, 8, 68, 128]) - layer.2.v_cache: torch.Size([1, 8, 68, 128]) - layer.3.k_cache: torch.Size([1, 8, 68, 128]) - layer.3.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.k_cache: torch.Size([1, 8, 68, 128]) - layer.4.v_cache: torch.Size([1, 8, 68, 128]) - layer.4.output: torch.Size([1, 68, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02721363 108.53914867 - layer.0.v_cache 0.00000029 0.00068831 - layer.1.k_cache 0.00369932 7.39369920 - layer.1.v_cache 0.00000081 0.00276004 - layer.2.k_cache 0.00117793 2.32072337 - layer.2.v_cache 0.00000112 0.00419039 - layer.3.k_cache 0.00140028 2.85580063 - layer.3.v_cache 0.00000224 0.00752708 - layer.4.k_cache 0.00315349 6.84938633 - layer.4.v_cache 0.00000301 0.01149102 - layer.4.output 0.00021496 0.26339845 - ------------------------------------------------------------------------------------- - TOTAL 0.00267943 9.21707206 - (elements=974,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 974848 -Total Bytes 16788 -BPFP 0.1378 bits/point -EBPFP 0.2755 equivalent bits/point -MSE 9.217072 ----------------------- -------------------------------------------------------- -Time: 0.471s Load: 0.006s, Pack+Encode: 0.195s, Decode+Unpack: 0.271s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 68, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 68, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.2171 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample656-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample656-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample66-layer4-item1.zst (81/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample66-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 136, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 136, 128) -Output shape: (1, 136, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.0.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.1.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.1.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.2.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.2.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.3.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.3.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.4.k_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.4.v_cache: torch.Size([1, 8, 136, 128]) -> torch.Size([1, 1, 136, 1024]) - layer.4.output: torch.Size([1, 136, 4096]) -> torch.Size([1, 1, 136, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,660B, BPFP=0.0954 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,700B, BPFP=0.0977 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,464B, BPFP=0.1415 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,012B, BPFP=0.1156 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,480B, BPFP=0.1425 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,132B, BPFP=0.1225 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,612B, BPFP=0.1500 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,220B, BPFP=0.1275 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,296B, BPFP=0.1319 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,212B, BPFP=0.1271 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 6,372B, BPFP=0.0915 -⌛️ [2/4] FRONTEND: Frontend time: 0.243s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 136, 128]) - layer.0.v_cache: torch.Size([1, 8, 136, 128]) - layer.1.k_cache: torch.Size([1, 8, 136, 128]) - layer.1.v_cache: torch.Size([1, 8, 136, 128]) - layer.2.k_cache: torch.Size([1, 8, 136, 128]) - layer.2.v_cache: torch.Size([1, 8, 136, 128]) - layer.3.k_cache: torch.Size([1, 8, 136, 128]) - layer.3.v_cache: torch.Size([1, 8, 136, 128]) - layer.4.k_cache: torch.Size([1, 8, 136, 128]) - layer.4.v_cache: torch.Size([1, 8, 136, 128]) - layer.4.output: torch.Size([1, 136, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.363s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 136, 128]) - layer.0.v_cache: torch.Size([1, 8, 136, 128]) - layer.1.k_cache: torch.Size([1, 8, 136, 128]) - layer.1.v_cache: torch.Size([1, 8, 136, 128]) - layer.2.k_cache: torch.Size([1, 8, 136, 128]) - layer.2.v_cache: torch.Size([1, 8, 136, 128]) - layer.3.k_cache: torch.Size([1, 8, 136, 128]) - layer.3.v_cache: torch.Size([1, 8, 136, 128]) - layer.4.k_cache: torch.Size([1, 8, 136, 128]) - layer.4.v_cache: torch.Size([1, 8, 136, 128]) - layer.4.output: torch.Size([1, 136, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02725438 73.67906997 - layer.0.v_cache 0.00000027 0.00067748 - layer.1.k_cache 0.00321745 6.38872932 - layer.1.v_cache 0.00000084 0.00262903 - layer.2.k_cache 0.00114365 2.14618234 - layer.2.v_cache 0.00000128 0.00410842 - layer.3.k_cache 0.00137918 2.59226047 - layer.3.v_cache 0.00000219 0.00721982 - layer.4.k_cache 0.00340721 6.57307883 - layer.4.v_cache 0.00000338 0.01151743 - layer.4.output 0.00019585 0.24677658 - ------------------------------------------------------------------------------------- - TOTAL 0.00265666 6.59946996 - (elements=1,949,696) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1949696 -Total Bytes 28160 -BPFP 0.1155 bits/point -EBPFP 0.2311 equivalent bits/point -MSE 6.599470 ----------------------- -------------------------------------------------------- -Time: 0.615s Load: 0.009s, Pack+Encode: 0.243s, Decode+Unpack: 0.363s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 136, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 136, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.5995 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample66-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample66-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample663-layer4-item1.zst (82/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample663-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 67, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 67, 128) -Output shape: (1, 67, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.0.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.1.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.1.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.2.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.2.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.3.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.3.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.4.k_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.4.v_cache: torch.Size([1, 8, 67, 128]) -> torch.Size([1, 1, 67, 1024]) - layer.4.output: torch.Size([1, 67, 4096]) -> torch.Size([1, 1, 67, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 968B, BPFP=0.1129 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,032B, BPFP=0.1203 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,360B, BPFP=0.1586 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,192B, BPFP=0.1390 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,392B, BPFP=0.1623 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,288B, BPFP=0.1502 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1679 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,324B, BPFP=0.1544 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,268B, BPFP=0.1479 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,284B, BPFP=0.1497 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,848B, BPFP=0.1122 -⌛️ [2/4] FRONTEND: Frontend time: 0.198s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 67, 128]) - layer.0.v_cache: torch.Size([1, 8, 67, 128]) - layer.1.k_cache: torch.Size([1, 8, 67, 128]) - layer.1.v_cache: torch.Size([1, 8, 67, 128]) - layer.2.k_cache: torch.Size([1, 8, 67, 128]) - layer.2.v_cache: torch.Size([1, 8, 67, 128]) - layer.3.k_cache: torch.Size([1, 8, 67, 128]) - layer.3.v_cache: torch.Size([1, 8, 67, 128]) - layer.4.k_cache: torch.Size([1, 8, 67, 128]) - layer.4.v_cache: torch.Size([1, 8, 67, 128]) - layer.4.output: torch.Size([1, 67, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.276s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 67, 128]) - layer.0.v_cache: torch.Size([1, 8, 67, 128]) - layer.1.k_cache: torch.Size([1, 8, 67, 128]) - layer.1.v_cache: torch.Size([1, 8, 67, 128]) - layer.2.k_cache: torch.Size([1, 8, 67, 128]) - layer.2.v_cache: torch.Size([1, 8, 67, 128]) - layer.3.k_cache: torch.Size([1, 8, 67, 128]) - layer.3.v_cache: torch.Size([1, 8, 67, 128]) - layer.4.k_cache: torch.Size([1, 8, 67, 128]) - layer.4.v_cache: torch.Size([1, 8, 67, 128]) - layer.4.output: torch.Size([1, 67, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02769006 193.28401936 - layer.0.v_cache 0.00000028 0.00070055 - layer.1.k_cache 0.00368886 7.65055460 - layer.1.v_cache 0.00000112 0.00293749 - layer.2.k_cache 0.00116499 2.44232565 - layer.2.v_cache 0.00000120 0.00466434 - layer.3.k_cache 0.00139456 2.97397283 - layer.3.v_cache 0.00000232 0.00806418 - layer.4.k_cache 0.00321227 7.31903441 - layer.4.v_cache 0.00000307 0.01209574 - layer.4.output 0.00021890 0.27548226 - ------------------------------------------------------------------------------------- - TOTAL 0.00271674 15.34287844 - (elements=960,512) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 960512 -Total Bytes 16396 -BPFP 0.1366 bits/point -EBPFP 0.2731 equivalent bits/point -MSE 15.342878 ----------------------- -------------------------------------------------------- -Time: 0.478s Load: 0.004s, Pack+Encode: 0.198s, Decode+Unpack: 0.276s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 67, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 67, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 15.3429 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample663-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample663-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample68-layer4-item1.zst (83/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample68-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 123, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 123, 128) -Output shape: (1, 123, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.0.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.1.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.1.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.2.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.2.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.3.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.3.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.4.k_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.4.v_cache: torch.Size([1, 8, 123, 128]) -> torch.Size([1, 1, 123, 1024]) - layer.4.output: torch.Size([1, 123, 4096]) -> torch.Size([1, 1, 123, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,048B, BPFP=0.0666 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,044B, BPFP=0.0663 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,384B, BPFP=0.0879 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,232B, BPFP=0.0783 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,452B, BPFP=0.0922 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,276B, BPFP=0.0810 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,496B, BPFP=0.0950 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,340B, BPFP=0.0851 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,356B, BPFP=0.0861 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,304B, BPFP=0.0828 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,668B, BPFP=0.0582 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 123, 128]) - layer.0.v_cache: torch.Size([1, 8, 123, 128]) - layer.1.k_cache: torch.Size([1, 8, 123, 128]) - layer.1.v_cache: torch.Size([1, 8, 123, 128]) - layer.2.k_cache: torch.Size([1, 8, 123, 128]) - layer.2.v_cache: torch.Size([1, 8, 123, 128]) - layer.3.k_cache: torch.Size([1, 8, 123, 128]) - layer.3.v_cache: torch.Size([1, 8, 123, 128]) - layer.4.k_cache: torch.Size([1, 8, 123, 128]) - layer.4.v_cache: torch.Size([1, 8, 123, 128]) - layer.4.output: torch.Size([1, 123, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.282s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 123, 128]) - layer.0.v_cache: torch.Size([1, 8, 123, 128]) - layer.1.k_cache: torch.Size([1, 8, 123, 128]) - layer.1.v_cache: torch.Size([1, 8, 123, 128]) - layer.2.k_cache: torch.Size([1, 8, 123, 128]) - layer.2.v_cache: torch.Size([1, 8, 123, 128]) - layer.3.k_cache: torch.Size([1, 8, 123, 128]) - layer.3.v_cache: torch.Size([1, 8, 123, 128]) - layer.4.k_cache: torch.Size([1, 8, 123, 128]) - layer.4.v_cache: torch.Size([1, 8, 123, 128]) - layer.4.output: torch.Size([1, 123, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02602563 66.84479008 - layer.0.v_cache 0.00000027 0.00067726 - layer.1.k_cache 0.00336899 6.52729462 - layer.1.v_cache 0.00000079 0.00253096 - layer.2.k_cache 0.00113470 2.12535455 - layer.2.v_cache 0.00000113 0.00399075 - layer.3.k_cache 0.00144113 2.58731129 - layer.3.v_cache 0.00000264 0.00728491 - layer.4.k_cache 0.00356668 6.84675859 - layer.4.v_cache 0.00000295 0.01108398 - layer.4.output 0.00019508 0.25357907 - ------------------------------------------------------------------------------------- - TOTAL 0.00259466 6.14081380 - (elements=1,763,328) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1763328 -Total Bytes 16600 -BPFP 0.0753 bits/point -EBPFP 0.1506 equivalent bits/point -MSE 6.140814 ----------------------- -------------------------------------------------------- -Time: 0.495s Load: 0.008s, Pack+Encode: 0.204s, Decode+Unpack: 0.282s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 123, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 123, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.1408 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample68-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample68-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample7-layer4-item1.zst (84/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample7-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.009s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 157, 128) -Output shape: (1, 157, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) -> torch.Size([1, 1, 157, 1024]) - layer.4.output: torch.Size([1, 157, 4096]) -> torch.Size([1, 1, 157, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,452B, BPFP=0.0723 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,472B, BPFP=0.0732 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,940B, BPFP=0.0965 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,616B, BPFP=0.0804 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,068B, BPFP=0.1029 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,752B, BPFP=0.0872 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,184B, BPFP=0.1087 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,864B, BPFP=0.0928 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,844B, BPFP=0.0918 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,776B, BPFP=0.0884 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 5,868B, BPFP=0.0730 -⌛️ [2/4] FRONTEND: Frontend time: 0.252s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.379s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 157, 128]) - layer.0.v_cache: torch.Size([1, 8, 157, 128]) - layer.1.k_cache: torch.Size([1, 8, 157, 128]) - layer.1.v_cache: torch.Size([1, 8, 157, 128]) - layer.2.k_cache: torch.Size([1, 8, 157, 128]) - layer.2.v_cache: torch.Size([1, 8, 157, 128]) - layer.3.k_cache: torch.Size([1, 8, 157, 128]) - layer.3.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.k_cache: torch.Size([1, 8, 157, 128]) - layer.4.v_cache: torch.Size([1, 8, 157, 128]) - layer.4.output: torch.Size([1, 157, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02423319 64.38334743 - layer.0.v_cache 0.00000027 0.00066051 - layer.1.k_cache 0.00309445 6.26822191 - layer.1.v_cache 0.00000088 0.00263598 - layer.2.k_cache 0.00117908 2.10251107 - layer.2.v_cache 0.00000110 0.00400049 - layer.3.k_cache 0.00138553 2.52440454 - layer.3.v_cache 0.00000215 0.00712034 - layer.4.k_cache 0.00349636 6.42597096 - layer.4.v_cache 0.00000310 0.01111465 - layer.4.output 0.00020206 0.21642129 - ------------------------------------------------------------------------------------- - TOTAL 0.00244317 5.89969093 - (elements=2,250,752) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2250752 -Total Bytes 23836 -BPFP 0.0847 bits/point -EBPFP 0.1694 equivalent bits/point -MSE 5.899691 ----------------------- -------------------------------------------------------- -Time: 0.641s Load: 0.009s, Pack+Encode: 0.252s, Decode+Unpack: 0.379s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 157, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 157, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.8997 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample7-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample7-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample736-layer4-item1.zst (85/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample736-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 75, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 75, 128) -Output shape: (1, 75, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.0.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.1.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.1.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.2.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.2.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.3.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.3.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.4.k_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.4.v_cache: torch.Size([1, 8, 75, 128]) -> torch.Size([1, 1, 75, 1024]) - layer.4.output: torch.Size([1, 75, 4096]) -> torch.Size([1, 1, 75, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 956B, BPFP=0.0996 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,084B, BPFP=0.1129 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,332B, BPFP=0.1388 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,132B, BPFP=0.1179 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,388B, BPFP=0.1446 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,216B, BPFP=0.1267 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1500 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,312B, BPFP=0.1367 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,224B, BPFP=0.1275 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,256B, BPFP=0.1308 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,032B, BPFP=0.1050 -⌛️ [2/4] FRONTEND: Frontend time: 0.206s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 75, 128]) - layer.0.v_cache: torch.Size([1, 8, 75, 128]) - layer.1.k_cache: torch.Size([1, 8, 75, 128]) - layer.1.v_cache: torch.Size([1, 8, 75, 128]) - layer.2.k_cache: torch.Size([1, 8, 75, 128]) - layer.2.v_cache: torch.Size([1, 8, 75, 128]) - layer.3.k_cache: torch.Size([1, 8, 75, 128]) - layer.3.v_cache: torch.Size([1, 8, 75, 128]) - layer.4.k_cache: torch.Size([1, 8, 75, 128]) - layer.4.v_cache: torch.Size([1, 8, 75, 128]) - layer.4.output: torch.Size([1, 75, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.281s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 75, 128]) - layer.0.v_cache: torch.Size([1, 8, 75, 128]) - layer.1.k_cache: torch.Size([1, 8, 75, 128]) - layer.1.v_cache: torch.Size([1, 8, 75, 128]) - layer.2.k_cache: torch.Size([1, 8, 75, 128]) - layer.2.v_cache: torch.Size([1, 8, 75, 128]) - layer.3.k_cache: torch.Size([1, 8, 75, 128]) - layer.3.v_cache: torch.Size([1, 8, 75, 128]) - layer.4.k_cache: torch.Size([1, 8, 75, 128]) - layer.4.v_cache: torch.Size([1, 8, 75, 128]) - layer.4.output: torch.Size([1, 75, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02489581 108.07610026 - layer.0.v_cache 0.00000028 0.00071368 - layer.1.k_cache 0.00341811 7.05050212 - layer.1.v_cache 0.00000080 0.00273420 - layer.2.k_cache 0.00110890 2.30075358 - layer.2.v_cache 0.00000112 0.00411107 - layer.3.k_cache 0.00134674 2.83381694 - layer.3.v_cache 0.00000209 0.00713957 - layer.4.k_cache 0.00330252 6.83343424 - layer.4.v_cache 0.00000296 0.01145038 - layer.4.output 0.00020017 0.26650993 - ------------------------------------------------------------------------------------- - TOTAL 0.00249143 9.15619970 - (elements=1,075,200) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1075200 -Total Bytes 16372 -BPFP 0.1218 bits/point -EBPFP 0.2436 equivalent bits/point -MSE 9.156200 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.005s, Pack+Encode: 0.206s, Decode+Unpack: 0.281s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 75, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 75, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.1562 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample736-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample736-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample74-layer4-item1.zst (86/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample74-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 122, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 122, 128) -Output shape: (1, 122, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.0.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.1.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.1.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.2.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.2.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.3.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.3.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.4.k_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.4.v_cache: torch.Size([1, 8, 122, 128]) -> torch.Size([1, 1, 122, 1024]) - layer.4.output: torch.Size([1, 122, 4096]) -> torch.Size([1, 1, 122, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,072B, BPFP=0.0686 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,160B, BPFP=0.0743 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,456B, BPFP=0.0932 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,384B, BPFP=0.0886 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,520B, BPFP=0.0973 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,424B, BPFP=0.0912 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,576B, BPFP=0.1009 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,464B, BPFP=0.0938 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,440B, BPFP=0.0922 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,428B, BPFP=0.0914 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,012B, BPFP=0.0642 -⌛️ [2/4] FRONTEND: Frontend time: 0.198s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 122, 128]) - layer.0.v_cache: torch.Size([1, 8, 122, 128]) - layer.1.k_cache: torch.Size([1, 8, 122, 128]) - layer.1.v_cache: torch.Size([1, 8, 122, 128]) - layer.2.k_cache: torch.Size([1, 8, 122, 128]) - layer.2.v_cache: torch.Size([1, 8, 122, 128]) - layer.3.k_cache: torch.Size([1, 8, 122, 128]) - layer.3.v_cache: torch.Size([1, 8, 122, 128]) - layer.4.k_cache: torch.Size([1, 8, 122, 128]) - layer.4.v_cache: torch.Size([1, 8, 122, 128]) - layer.4.output: torch.Size([1, 122, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.273s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 122, 128]) - layer.0.v_cache: torch.Size([1, 8, 122, 128]) - layer.1.k_cache: torch.Size([1, 8, 122, 128]) - layer.1.v_cache: torch.Size([1, 8, 122, 128]) - layer.2.k_cache: torch.Size([1, 8, 122, 128]) - layer.2.v_cache: torch.Size([1, 8, 122, 128]) - layer.3.k_cache: torch.Size([1, 8, 122, 128]) - layer.3.v_cache: torch.Size([1, 8, 122, 128]) - layer.4.k_cache: torch.Size([1, 8, 122, 128]) - layer.4.v_cache: torch.Size([1, 8, 122, 128]) - layer.4.output: torch.Size([1, 122, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02525679 100.32233767 - layer.0.v_cache 0.00000028 0.00070462 - layer.1.k_cache 0.00316206 6.86338731 - layer.1.v_cache 0.00000097 0.00286734 - layer.2.k_cache 0.00113900 2.22147607 - layer.2.v_cache 0.00000117 0.00437395 - layer.3.k_cache 0.00136642 2.70158586 - layer.3.v_cache 0.00000221 0.00746566 - layer.4.k_cache 0.00339647 6.71021658 - layer.4.v_cache 0.00000316 0.01174224 - layer.4.output 0.00022994 0.24135000 - ------------------------------------------------------------------------------------- - TOTAL 0.00251773 8.55796838 - (elements=1,748,992) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1748992 -Total Bytes 17936 -BPFP 0.0820 bits/point -EBPFP 0.1641 equivalent bits/point -MSE 8.557968 ----------------------- -------------------------------------------------------- -Time: 0.479s Load: 0.008s, Pack+Encode: 0.198s, Decode+Unpack: 0.273s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 122, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 122, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.5580 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample74-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample74-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample75-layer4-item1.zst (87/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample75-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 142, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.008s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 142, 128) -Output shape: (1, 142, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.0.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.1.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.1.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.2.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.2.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.3.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.3.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.4.k_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.4.v_cache: torch.Size([1, 8, 142, 128]) -> torch.Size([1, 1, 142, 1024]) - layer.4.output: torch.Size([1, 142, 4096]) -> torch.Size([1, 1, 142, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,608B, BPFP=0.0885 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,764B, BPFP=0.0971 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,268B, BPFP=0.1248 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,100B, BPFP=0.1155 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 2,412B, BPFP=0.1327 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,184B, BPFP=0.1202 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 2,528B, BPFP=0.1391 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,340B, BPFP=0.1287 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,212B, BPFP=0.1217 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,248B, BPFP=0.1237 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 7,172B, BPFP=0.0986 -⌛️ [2/4] FRONTEND: Frontend time: 0.245s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 142, 128]) - layer.0.v_cache: torch.Size([1, 8, 142, 128]) - layer.1.k_cache: torch.Size([1, 8, 142, 128]) - layer.1.v_cache: torch.Size([1, 8, 142, 128]) - layer.2.k_cache: torch.Size([1, 8, 142, 128]) - layer.2.v_cache: torch.Size([1, 8, 142, 128]) - layer.3.k_cache: torch.Size([1, 8, 142, 128]) - layer.3.v_cache: torch.Size([1, 8, 142, 128]) - layer.4.k_cache: torch.Size([1, 8, 142, 128]) - layer.4.v_cache: torch.Size([1, 8, 142, 128]) - layer.4.output: torch.Size([1, 142, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.367s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 142, 128]) - layer.0.v_cache: torch.Size([1, 8, 142, 128]) - layer.1.k_cache: torch.Size([1, 8, 142, 128]) - layer.1.v_cache: torch.Size([1, 8, 142, 128]) - layer.2.k_cache: torch.Size([1, 8, 142, 128]) - layer.2.v_cache: torch.Size([1, 8, 142, 128]) - layer.3.k_cache: torch.Size([1, 8, 142, 128]) - layer.3.v_cache: torch.Size([1, 8, 142, 128]) - layer.4.k_cache: torch.Size([1, 8, 142, 128]) - layer.4.v_cache: torch.Size([1, 8, 142, 128]) - layer.4.output: torch.Size([1, 142, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02301751 71.46714761 - layer.0.v_cache 0.00000029 0.00068827 - layer.1.k_cache 0.00315333 6.34375645 - layer.1.v_cache 0.00000086 0.00266630 - layer.2.k_cache 0.00112591 2.11272097 - layer.2.v_cache 0.00000130 0.00412068 - layer.3.k_cache 0.00130425 2.60022467 - layer.3.v_cache 0.00000237 0.00770633 - layer.4.k_cache 0.00348660 6.68920426 - layer.4.v_cache 0.00000314 0.01157985 - layer.4.output 0.00018111 0.24823772 - ------------------------------------------------------------------------------------- - TOTAL 0.00234429 6.44519759 - (elements=2,035,712) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 2035712 -Total Bytes 28836 -BPFP 0.1133 bits/point -EBPFP 0.2266 equivalent bits/point -MSE 6.445198 ----------------------- -------------------------------------------------------- -Time: 0.620s Load: 0.008s, Pack+Encode: 0.245s, Decode+Unpack: 0.367s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 142, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 142, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.4452 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample75-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample75-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample777-layer4-item1.zst (88/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample777-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 74, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 74, 128) -Output shape: (1, 74, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) -> torch.Size([1, 1, 74, 1024]) - layer.4.output: torch.Size([1, 74, 4096]) -> torch.Size([1, 1, 74, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 984B, BPFP=0.1039 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,024B, BPFP=0.1081 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,344B, BPFP=0.1419 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,164B, BPFP=0.1229 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,404B, BPFP=0.1482 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,264B, BPFP=0.1334 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,440B, BPFP=0.1520 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,320B, BPFP=0.1394 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,228B, BPFP=0.1296 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,244B, BPFP=0.1313 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,964B, BPFP=0.1046 -⌛️ [2/4] FRONTEND: Frontend time: 0.197s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.output: torch.Size([1, 74, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.270s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 74, 128]) - layer.0.v_cache: torch.Size([1, 8, 74, 128]) - layer.1.k_cache: torch.Size([1, 8, 74, 128]) - layer.1.v_cache: torch.Size([1, 8, 74, 128]) - layer.2.k_cache: torch.Size([1, 8, 74, 128]) - layer.2.v_cache: torch.Size([1, 8, 74, 128]) - layer.3.k_cache: torch.Size([1, 8, 74, 128]) - layer.3.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.k_cache: torch.Size([1, 8, 74, 128]) - layer.4.v_cache: torch.Size([1, 8, 74, 128]) - layer.4.output: torch.Size([1, 74, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02555943 100.97178526 - layer.0.v_cache 0.00000028 0.00068737 - layer.1.k_cache 0.00343224 7.03752507 - layer.1.v_cache 0.00000079 0.00264254 - layer.2.k_cache 0.00111603 2.25002351 - layer.2.v_cache 0.00000116 0.00406979 - layer.3.k_cache 0.00133047 2.79968262 - layer.3.v_cache 0.00000218 0.00717913 - layer.4.k_cache 0.00334566 6.83134873 - layer.4.v_cache 0.00000301 0.01109373 - layer.4.output 0.00024685 0.25053983 - ------------------------------------------------------------------------------------- - TOTAL 0.00255562 8.63701408 - (elements=1,060,864) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1060864 -Total Bytes 16380 -BPFP 0.1235 bits/point -EBPFP 0.2470 equivalent bits/point -MSE 8.637014 ----------------------- -------------------------------------------------------- -Time: 0.473s Load: 0.005s, Pack+Encode: 0.197s, Decode+Unpack: 0.270s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 74, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 74, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.6370 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample777-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample777-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample778-layer4-item1.zst (89/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample778-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 61, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 61, 128) -Output shape: (1, 61, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.0.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.1.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.1.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.2.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.2.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.3.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.3.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.4.k_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.4.v_cache: torch.Size([1, 8, 61, 128]) -> torch.Size([1, 1, 61, 1024]) - layer.4.output: torch.Size([1, 61, 4096]) -> torch.Size([1, 1, 61, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 544B, BPFP=0.0697 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 588B, BPFP=0.0753 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 684B, BPFP=0.0876 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 636B, BPFP=0.0815 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 716B, BPFP=0.0917 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 664B, BPFP=0.0850 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 740B, BPFP=0.0948 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 676B, BPFP=0.0866 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 660B, BPFP=0.0845 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 660B, BPFP=0.0845 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 2,144B, BPFP=0.0686 -⌛️ [2/4] FRONTEND: Frontend time: 0.146s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 61, 128]) - layer.0.v_cache: torch.Size([1, 8, 61, 128]) - layer.1.k_cache: torch.Size([1, 8, 61, 128]) - layer.1.v_cache: torch.Size([1, 8, 61, 128]) - layer.2.k_cache: torch.Size([1, 8, 61, 128]) - layer.2.v_cache: torch.Size([1, 8, 61, 128]) - layer.3.k_cache: torch.Size([1, 8, 61, 128]) - layer.3.v_cache: torch.Size([1, 8, 61, 128]) - layer.4.k_cache: torch.Size([1, 8, 61, 128]) - layer.4.v_cache: torch.Size([1, 8, 61, 128]) - layer.4.output: torch.Size([1, 61, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.188s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 61, 128]) - layer.0.v_cache: torch.Size([1, 8, 61, 128]) - layer.1.k_cache: torch.Size([1, 8, 61, 128]) - layer.1.v_cache: torch.Size([1, 8, 61, 128]) - layer.2.k_cache: torch.Size([1, 8, 61, 128]) - layer.2.v_cache: torch.Size([1, 8, 61, 128]) - layer.3.k_cache: torch.Size([1, 8, 61, 128]) - layer.3.v_cache: torch.Size([1, 8, 61, 128]) - layer.4.k_cache: torch.Size([1, 8, 61, 128]) - layer.4.v_cache: torch.Size([1, 8, 61, 128]) - layer.4.output: torch.Size([1, 61, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02755715 63.39192975 - layer.0.v_cache 0.00000026 0.00062990 - layer.1.k_cache 0.00385674 6.59992256 - layer.1.v_cache 0.00000079 0.00278846 - layer.2.k_cache 0.00117960 2.14107326 - layer.2.v_cache 0.00000113 0.00419695 - layer.3.k_cache 0.00144752 2.57355062 - layer.3.v_cache 0.00000212 0.00696634 - layer.4.k_cache 0.00316257 6.05139861 - layer.4.v_cache 0.00000306 0.01119024 - layer.4.output 0.00025867 0.27831131 - ------------------------------------------------------------------------------------- - TOTAL 0.00273183 5.84977799 - (elements=874,496) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 874496 -Total Bytes 8712 -BPFP 0.0797 bits/point -EBPFP 0.1594 equivalent bits/point -MSE 5.849778 ----------------------- -------------------------------------------------------- -Time: 0.339s Load: 0.005s, Pack+Encode: 0.146s, Decode+Unpack: 0.188s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 61, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 61, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.8498 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample778-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample778-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample807-layer4-item1.zst (90/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample807-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 59, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 59, 128) -Output shape: (1, 59, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.output: torch.Size([1, 59, 4096]) -> torch.Size([1, 1, 59, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 556B, BPFP=0.0736 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 604B, BPFP=0.0800 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 772B, BPFP=0.1022 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 724B, BPFP=0.0959 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 796B, BPFP=0.1054 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 740B, BPFP=0.0980 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 816B, BPFP=0.1081 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 772B, BPFP=0.1022 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 756B, BPFP=0.1001 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 728B, BPFP=0.0964 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 2,256B, BPFP=0.0747 -⌛️ [2/4] FRONTEND: Frontend time: 0.147s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.output: torch.Size([1, 59, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.190s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.output: torch.Size([1, 59, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.03100202 116.60719015 - layer.0.v_cache 0.00000029 0.00074002 - layer.1.k_cache 0.00372036 7.62226376 - layer.1.v_cache 0.00000085 0.00306562 - layer.2.k_cache 0.00115281 2.41855841 - layer.2.v_cache 0.00000145 0.00471030 - layer.3.k_cache 0.00140016 2.89254399 - layer.3.v_cache 0.00000257 0.00831168 - layer.4.k_cache 0.00323529 7.15149598 - layer.4.v_cache 0.00000321 0.01241934 - layer.4.output 0.00024217 0.29812877 - ------------------------------------------------------------------------------------- - TOTAL 0.00296341 9.85098674 - (elements=845,824) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 845824 -Total Bytes 9520 -BPFP 0.0900 bits/point -EBPFP 0.1801 equivalent bits/point -MSE 9.850987 ----------------------- -------------------------------------------------------- -Time: 0.341s Load: 0.005s, Pack+Encode: 0.147s, Decode+Unpack: 0.190s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 59, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.8510 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample807-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample807-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample855-layer4-item1.zst (91/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample855-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 66, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 66, 128) -Output shape: (1, 66, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.0.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.1.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.1.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.2.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.2.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.3.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.3.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.4.k_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.4.v_cache: torch.Size([1, 8, 66, 128]) -> torch.Size([1, 1, 66, 1024]) - layer.4.output: torch.Size([1, 66, 4096]) -> torch.Size([1, 1, 66, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 924B, BPFP=0.1094 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 936B, BPFP=0.1108 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,332B, BPFP=0.1577 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,108B, BPFP=0.1312 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,376B, BPFP=0.1629 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,244B, BPFP=0.1473 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,424B, BPFP=0.1686 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,264B, BPFP=0.1496 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,240B, BPFP=0.1468 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,248B, BPFP=0.1477 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,624B, BPFP=0.1072 -⌛️ [2/4] FRONTEND: Frontend time: 0.197s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 66, 128]) - layer.0.v_cache: torch.Size([1, 8, 66, 128]) - layer.1.k_cache: torch.Size([1, 8, 66, 128]) - layer.1.v_cache: torch.Size([1, 8, 66, 128]) - layer.2.k_cache: torch.Size([1, 8, 66, 128]) - layer.2.v_cache: torch.Size([1, 8, 66, 128]) - layer.3.k_cache: torch.Size([1, 8, 66, 128]) - layer.3.v_cache: torch.Size([1, 8, 66, 128]) - layer.4.k_cache: torch.Size([1, 8, 66, 128]) - layer.4.v_cache: torch.Size([1, 8, 66, 128]) - layer.4.output: torch.Size([1, 66, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.270s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 66, 128]) - layer.0.v_cache: torch.Size([1, 8, 66, 128]) - layer.1.k_cache: torch.Size([1, 8, 66, 128]) - layer.1.v_cache: torch.Size([1, 8, 66, 128]) - layer.2.k_cache: torch.Size([1, 8, 66, 128]) - layer.2.v_cache: torch.Size([1, 8, 66, 128]) - layer.3.k_cache: torch.Size([1, 8, 66, 128]) - layer.3.v_cache: torch.Size([1, 8, 66, 128]) - layer.4.k_cache: torch.Size([1, 8, 66, 128]) - layer.4.v_cache: torch.Size([1, 8, 66, 128]) - layer.4.output: torch.Size([1, 66, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02815026 219.63559422 - layer.0.v_cache 0.00000028 0.00068952 - layer.1.k_cache 0.00365239 7.75250707 - layer.1.v_cache 0.00000074 0.00262635 - layer.2.k_cache 0.00118996 2.43306755 - layer.2.v_cache 0.00000101 0.00388069 - layer.3.k_cache 0.00141080 2.92936545 - layer.3.v_cache 0.00000191 0.00663749 - layer.4.k_cache 0.00329622 7.51569713 - layer.4.v_cache 0.00000294 0.01078349 - layer.4.output 0.00019151 0.30174799 - ------------------------------------------------------------------------------------- - TOTAL 0.00274804 17.24984578 - (elements=946,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 946176 -Total Bytes 15720 -BPFP 0.1329 bits/point -EBPFP 0.2658 equivalent bits/point -MSE 17.249846 ----------------------- -------------------------------------------------------- -Time: 0.472s Load: 0.005s, Pack+Encode: 0.197s, Decode+Unpack: 0.270s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 66, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 66, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 17.2498 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample855-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample855-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample859-layer4-item1.zst (92/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample859-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.005s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 73, 128) -Output shape: (1, 73, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) -> torch.Size([1, 1, 73, 1024]) - layer.4.output: torch.Size([1, 73, 4096]) -> torch.Size([1, 1, 73, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,000B, BPFP=0.1070 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,076B, BPFP=0.1152 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,340B, BPFP=0.1434 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,144B, BPFP=0.1224 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,408B, BPFP=0.1507 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,244B, BPFP=0.1331 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,444B, BPFP=0.1545 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,332B, BPFP=0.1426 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,248B, BPFP=0.1336 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,272B, BPFP=0.1361 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,172B, BPFP=0.1116 -⌛️ [2/4] FRONTEND: Frontend time: 0.197s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.279s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 73, 128]) - layer.0.v_cache: torch.Size([1, 8, 73, 128]) - layer.1.k_cache: torch.Size([1, 8, 73, 128]) - layer.1.v_cache: torch.Size([1, 8, 73, 128]) - layer.2.k_cache: torch.Size([1, 8, 73, 128]) - layer.2.v_cache: torch.Size([1, 8, 73, 128]) - layer.3.k_cache: torch.Size([1, 8, 73, 128]) - layer.3.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.k_cache: torch.Size([1, 8, 73, 128]) - layer.4.v_cache: torch.Size([1, 8, 73, 128]) - layer.4.output: torch.Size([1, 73, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02483096 95.71182042 - layer.0.v_cache 0.00000028 0.00067291 - layer.1.k_cache 0.00355597 7.08094599 - layer.1.v_cache 0.00000080 0.00249667 - layer.2.k_cache 0.00115478 2.25829117 - layer.2.v_cache 0.00000105 0.00382563 - layer.3.k_cache 0.00135922 2.75102192 - layer.3.v_cache 0.00000200 0.00662393 - layer.4.k_cache 0.00333109 6.69432131 - layer.4.v_cache 0.00000279 0.01041623 - layer.4.output 0.00023199 0.24906258 - ------------------------------------------------------------------------------------- - TOTAL 0.00251192 8.25119189 - (elements=1,046,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1046528 -Total Bytes 16680 -BPFP 0.1275 bits/point -EBPFP 0.2550 equivalent bits/point -MSE 8.251192 ----------------------- -------------------------------------------------------- -Time: 0.481s Load: 0.005s, Pack+Encode: 0.197s, Decode+Unpack: 0.279s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 73, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 73, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.2512 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample859-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample859-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample86-layer4-item1.zst (93/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample86-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 121, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 121, 128) -Output shape: (1, 121, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.0.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.1.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.1.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.2.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.2.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.3.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.3.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.4.k_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.4.v_cache: torch.Size([1, 8, 121, 128]) -> torch.Size([1, 1, 121, 1024]) - layer.4.output: torch.Size([1, 121, 4096]) -> torch.Size([1, 1, 121, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,104B, BPFP=0.0713 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,160B, BPFP=0.0749 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,532B, BPFP=0.0989 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,380B, BPFP=0.0891 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,580B, BPFP=0.1020 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,460B, BPFP=0.0943 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,624B, BPFP=0.1049 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,536B, BPFP=0.0992 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,488B, BPFP=0.0961 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,472B, BPFP=0.0950 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,988B, BPFP=0.0644 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 121, 128]) - layer.0.v_cache: torch.Size([1, 8, 121, 128]) - layer.1.k_cache: torch.Size([1, 8, 121, 128]) - layer.1.v_cache: torch.Size([1, 8, 121, 128]) - layer.2.k_cache: torch.Size([1, 8, 121, 128]) - layer.2.v_cache: torch.Size([1, 8, 121, 128]) - layer.3.k_cache: torch.Size([1, 8, 121, 128]) - layer.3.v_cache: torch.Size([1, 8, 121, 128]) - layer.4.k_cache: torch.Size([1, 8, 121, 128]) - layer.4.v_cache: torch.Size([1, 8, 121, 128]) - layer.4.output: torch.Size([1, 121, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 121, 128]) - layer.0.v_cache: torch.Size([1, 8, 121, 128]) - layer.1.k_cache: torch.Size([1, 8, 121, 128]) - layer.1.v_cache: torch.Size([1, 8, 121, 128]) - layer.2.k_cache: torch.Size([1, 8, 121, 128]) - layer.2.v_cache: torch.Size([1, 8, 121, 128]) - layer.3.k_cache: torch.Size([1, 8, 121, 128]) - layer.3.v_cache: torch.Size([1, 8, 121, 128]) - layer.4.k_cache: torch.Size([1, 8, 121, 128]) - layer.4.v_cache: torch.Size([1, 8, 121, 128]) - layer.4.output: torch.Size([1, 121, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02495229 179.70317342 - layer.0.v_cache 0.00000028 0.00071082 - layer.1.k_cache 0.00332455 7.14020229 - layer.1.v_cache 0.00000080 0.00275234 - layer.2.k_cache 0.00116246 2.26882329 - layer.2.v_cache 0.00000109 0.00407748 - layer.3.k_cache 0.00134746 2.75691463 - layer.3.v_cache 0.00000209 0.00732107 - layer.4.k_cache 0.00345892 7.41064554 - layer.4.v_cache 0.00000309 0.01179156 - layer.4.output 0.00018371 0.25627177 - ------------------------------------------------------------------------------------- - TOTAL 0.00249913 14.30939282 - (elements=1,734,656) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1734656 -Total Bytes 18324 -BPFP 0.0845 bits/point -EBPFP 0.1690 equivalent bits/point -MSE 14.309393 ----------------------- -------------------------------------------------------- -Time: 0.495s Load: 0.007s, Pack+Encode: 0.204s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 121, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 121, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 14.3094 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample86-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample86-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample91-layer4-item1.zst (94/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample91-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 113, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.006s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 113, 128) -Output shape: (1, 113, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) -> torch.Size([1, 1, 113, 1024]) - layer.4.output: torch.Size([1, 113, 4096]) -> torch.Size([1, 1, 113, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,084B, BPFP=0.0749 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,136B, BPFP=0.0785 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,540B, BPFP=0.1065 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,348B, BPFP=0.0932 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,568B, BPFP=0.1084 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,484B, BPFP=0.1026 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,656B, BPFP=0.1145 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,500B, BPFP=0.1037 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,476B, BPFP=0.1020 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,452B, BPFP=0.1004 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 4,100B, BPFP=0.0709 -⌛️ [2/4] FRONTEND: Frontend time: 0.212s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.output: torch.Size([1, 113, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.288s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 113, 128]) - layer.0.v_cache: torch.Size([1, 8, 113, 128]) - layer.1.k_cache: torch.Size([1, 8, 113, 128]) - layer.1.v_cache: torch.Size([1, 8, 113, 128]) - layer.2.k_cache: torch.Size([1, 8, 113, 128]) - layer.2.v_cache: torch.Size([1, 8, 113, 128]) - layer.3.k_cache: torch.Size([1, 8, 113, 128]) - layer.3.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.k_cache: torch.Size([1, 8, 113, 128]) - layer.4.v_cache: torch.Size([1, 8, 113, 128]) - layer.4.output: torch.Size([1, 113, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02530831 84.58196211 - layer.0.v_cache 0.00000029 0.00067790 - layer.1.k_cache 0.00330439 6.43354514 - layer.1.v_cache 0.00000076 0.00254237 - layer.2.k_cache 0.00114562 2.13918203 - layer.2.v_cache 0.00000109 0.00392935 - layer.3.k_cache 0.00136171 2.61316161 - layer.3.v_cache 0.00000227 0.00739658 - layer.4.k_cache 0.00342222 6.42531241 - layer.4.v_cache 0.00000313 0.01150047 - layer.4.output 0.00018801 0.22821383 - ------------------------------------------------------------------------------------- - TOTAL 0.00252156 7.36657609 - (elements=1,619,968) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1619968 -Total Bytes 18344 -BPFP 0.0906 bits/point -EBPFP 0.1812 equivalent bits/point -MSE 7.366576 ----------------------- -------------------------------------------------------- -Time: 0.505s Load: 0.006s, Pack+Encode: 0.212s, Decode+Unpack: 0.288s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 113, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 113, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.3666 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample91-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample91-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample92-layer4-item1.zst (95/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample92-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 110, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 110, 128) -Output shape: (1, 110, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.0.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.1.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.1.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.2.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.2.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.3.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.3.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.4.k_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.4.v_cache: torch.Size([1, 8, 110, 128]) -> torch.Size([1, 1, 110, 1024]) - layer.4.output: torch.Size([1, 110, 4096]) -> torch.Size([1, 1, 110, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,112B, BPFP=0.0790 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,172B, BPFP=0.0832 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,520B, BPFP=0.1080 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,424B, BPFP=0.1011 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,548B, BPFP=0.1099 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,464B, BPFP=0.1040 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,588B, BPFP=0.1128 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,508B, BPFP=0.1071 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,456B, BPFP=0.1034 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,472B, BPFP=0.1045 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,892B, BPFP=0.0691 -⌛️ [2/4] FRONTEND: Frontend time: 0.213s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 110, 128]) - layer.0.v_cache: torch.Size([1, 8, 110, 128]) - layer.1.k_cache: torch.Size([1, 8, 110, 128]) - layer.1.v_cache: torch.Size([1, 8, 110, 128]) - layer.2.k_cache: torch.Size([1, 8, 110, 128]) - layer.2.v_cache: torch.Size([1, 8, 110, 128]) - layer.3.k_cache: torch.Size([1, 8, 110, 128]) - layer.3.v_cache: torch.Size([1, 8, 110, 128]) - layer.4.k_cache: torch.Size([1, 8, 110, 128]) - layer.4.v_cache: torch.Size([1, 8, 110, 128]) - layer.4.output: torch.Size([1, 110, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 110, 128]) - layer.0.v_cache: torch.Size([1, 8, 110, 128]) - layer.1.k_cache: torch.Size([1, 8, 110, 128]) - layer.1.v_cache: torch.Size([1, 8, 110, 128]) - layer.2.k_cache: torch.Size([1, 8, 110, 128]) - layer.2.v_cache: torch.Size([1, 8, 110, 128]) - layer.3.k_cache: torch.Size([1, 8, 110, 128]) - layer.3.v_cache: torch.Size([1, 8, 110, 128]) - layer.4.k_cache: torch.Size([1, 8, 110, 128]) - layer.4.v_cache: torch.Size([1, 8, 110, 128]) - layer.4.output: torch.Size([1, 110, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02547969 115.57962536 - layer.0.v_cache 0.00000028 0.00068481 - layer.1.k_cache 0.00323236 6.71134533 - layer.1.v_cache 0.00000082 0.00260852 - layer.2.k_cache 0.00114874 2.26089658 - layer.2.v_cache 0.00000131 0.00418129 - layer.3.k_cache 0.00138397 2.72541726 - layer.3.v_cache 0.00000212 0.00707982 - layer.4.k_cache 0.00346958 6.97403786 - layer.4.v_cache 0.00000311 0.01142503 - layer.4.output 0.00021756 0.25142548 - ------------------------------------------------------------------------------------- - TOTAL 0.00254230 9.66307170 - (elements=1,576,960) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1576960 -Total Bytes 18156 -BPFP 0.0921 bits/point -EBPFP 0.1842 equivalent bits/point -MSE 9.663072 ----------------------- -------------------------------------------------------- -Time: 0.504s Load: 0.007s, Pack+Encode: 0.213s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 110, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 110, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.6631 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample92-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample92-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample925-layer4-item1.zst (96/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample925-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 65, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 65, 128) -Output shape: (1, 65, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.0.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.1.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.1.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.2.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.2.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.3.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.3.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.4.k_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.4.v_cache: torch.Size([1, 8, 65, 128]) -> torch.Size([1, 1, 65, 1024]) - layer.4.output: torch.Size([1, 65, 4096]) -> torch.Size([1, 1, 65, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 924B, BPFP=0.1111 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 956B, BPFP=0.1149 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,336B, BPFP=0.1606 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,092B, BPFP=0.1313 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,344B, BPFP=0.1615 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,208B, BPFP=0.1452 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,396B, BPFP=0.1678 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,288B, BPFP=0.1548 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,228B, BPFP=0.1476 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,244B, BPFP=0.1495 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,780B, BPFP=0.1136 -⌛️ [2/4] FRONTEND: Frontend time: 0.204s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 65, 128]) - layer.0.v_cache: torch.Size([1, 8, 65, 128]) - layer.1.k_cache: torch.Size([1, 8, 65, 128]) - layer.1.v_cache: torch.Size([1, 8, 65, 128]) - layer.2.k_cache: torch.Size([1, 8, 65, 128]) - layer.2.v_cache: torch.Size([1, 8, 65, 128]) - layer.3.k_cache: torch.Size([1, 8, 65, 128]) - layer.3.v_cache: torch.Size([1, 8, 65, 128]) - layer.4.k_cache: torch.Size([1, 8, 65, 128]) - layer.4.v_cache: torch.Size([1, 8, 65, 128]) - layer.4.output: torch.Size([1, 65, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.284s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 65, 128]) - layer.0.v_cache: torch.Size([1, 8, 65, 128]) - layer.1.k_cache: torch.Size([1, 8, 65, 128]) - layer.1.v_cache: torch.Size([1, 8, 65, 128]) - layer.2.k_cache: torch.Size([1, 8, 65, 128]) - layer.2.v_cache: torch.Size([1, 8, 65, 128]) - layer.3.k_cache: torch.Size([1, 8, 65, 128]) - layer.3.v_cache: torch.Size([1, 8, 65, 128]) - layer.4.k_cache: torch.Size([1, 8, 65, 128]) - layer.4.v_cache: torch.Size([1, 8, 65, 128]) - layer.4.output: torch.Size([1, 65, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02613364 122.90077374 - layer.0.v_cache 0.00000028 0.00068207 - layer.1.k_cache 0.00351263 7.13489004 - layer.1.v_cache 0.00000086 0.00254324 - layer.2.k_cache 0.00113140 2.30734347 - layer.2.v_cache 0.00000103 0.00377396 - layer.3.k_cache 0.00134002 2.87169448 - layer.3.v_cache 0.00000212 0.00714943 - layer.4.k_cache 0.00331905 6.76536208 - layer.4.v_cache 0.00000293 0.01110973 - layer.4.output 0.00019456 0.26540504 - ------------------------------------------------------------------------------------- - TOTAL 0.00258730 10.21906731 - (elements=931,840) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 931840 -Total Bytes 15796 -BPFP 0.1356 bits/point -EBPFP 0.2712 equivalent bits/point -MSE 10.219067 ----------------------- -------------------------------------------------------- -Time: 0.493s Load: 0.004s, Pack+Encode: 0.204s, Decode+Unpack: 0.284s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 65, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 65, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.2191 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample925-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample925-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample95-layer4-item1.zst (97/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample95-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 116, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 116, 128) -Output shape: (1, 116, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.0.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.1.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.1.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.2.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.2.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.3.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.3.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.4.k_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.4.v_cache: torch.Size([1, 8, 116, 128]) -> torch.Size([1, 1, 116, 1024]) - layer.4.output: torch.Size([1, 116, 4096]) -> torch.Size([1, 1, 116, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,100B, BPFP=0.0741 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,092B, BPFP=0.0735 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,564B, BPFP=0.1053 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,340B, BPFP=0.0902 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,648B, BPFP=0.1110 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,496B, BPFP=0.1008 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,648B, BPFP=0.1110 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,556B, BPFP=0.1048 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,496B, BPFP=0.1008 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,548B, BPFP=0.1043 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,868B, BPFP=0.0651 -⌛️ [2/4] FRONTEND: Frontend time: 0.206s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 116, 128]) - layer.0.v_cache: torch.Size([1, 8, 116, 128]) - layer.1.k_cache: torch.Size([1, 8, 116, 128]) - layer.1.v_cache: torch.Size([1, 8, 116, 128]) - layer.2.k_cache: torch.Size([1, 8, 116, 128]) - layer.2.v_cache: torch.Size([1, 8, 116, 128]) - layer.3.k_cache: torch.Size([1, 8, 116, 128]) - layer.3.v_cache: torch.Size([1, 8, 116, 128]) - layer.4.k_cache: torch.Size([1, 8, 116, 128]) - layer.4.v_cache: torch.Size([1, 8, 116, 128]) - layer.4.output: torch.Size([1, 116, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.283s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 116, 128]) - layer.0.v_cache: torch.Size([1, 8, 116, 128]) - layer.1.k_cache: torch.Size([1, 8, 116, 128]) - layer.1.v_cache: torch.Size([1, 8, 116, 128]) - layer.2.k_cache: torch.Size([1, 8, 116, 128]) - layer.2.v_cache: torch.Size([1, 8, 116, 128]) - layer.3.k_cache: torch.Size([1, 8, 116, 128]) - layer.3.v_cache: torch.Size([1, 8, 116, 128]) - layer.4.k_cache: torch.Size([1, 8, 116, 128]) - layer.4.v_cache: torch.Size([1, 8, 116, 128]) - layer.4.output: torch.Size([1, 116, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02494503 134.72770744 - layer.0.v_cache 0.00000027 0.00067113 - layer.1.k_cache 0.00326550 6.71618758 - layer.1.v_cache 0.00000084 0.00269181 - layer.2.k_cache 0.00114506 2.18720929 - layer.2.v_cache 0.00000107 0.00403653 - layer.3.k_cache 0.00140586 2.70639827 - layer.3.v_cache 0.00000237 0.00727058 - layer.4.k_cache 0.00329219 6.58515036 - layer.4.v_cache 0.00000320 0.01177644 - layer.4.output 0.00021339 0.23720684 - ------------------------------------------------------------------------------------- - TOTAL 0.00249392 10.99270906 - (elements=1,662,976) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1662976 -Total Bytes 18356 -BPFP 0.0883 bits/point -EBPFP 0.1766 equivalent bits/point -MSE 10.992709 ----------------------- -------------------------------------------------------- -Time: 0.496s Load: 0.007s, Pack+Encode: 0.206s, Decode+Unpack: 0.283s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 116, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 116, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 10.9927 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample95-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample95-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample967-layer4-item1.zst (98/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample967-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 59, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 59, 128) -Output shape: (1, 59, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) -> torch.Size([1, 1, 59, 1024]) - layer.4.output: torch.Size([1, 59, 4096]) -> torch.Size([1, 1, 59, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 564B, BPFP=0.0747 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 576B, BPFP=0.0763 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 780B, BPFP=0.1033 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 664B, BPFP=0.0879 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 808B, BPFP=0.1070 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 720B, BPFP=0.0953 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 808B, BPFP=0.1070 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 748B, BPFP=0.0990 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 756B, BPFP=0.1001 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 712B, BPFP=0.0943 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 2,300B, BPFP=0.0761 -⌛️ [2/4] FRONTEND: Frontend time: 0.155s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.output: torch.Size([1, 59, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.190s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 59, 128]) - layer.0.v_cache: torch.Size([1, 8, 59, 128]) - layer.1.k_cache: torch.Size([1, 8, 59, 128]) - layer.1.v_cache: torch.Size([1, 8, 59, 128]) - layer.2.k_cache: torch.Size([1, 8, 59, 128]) - layer.2.v_cache: torch.Size([1, 8, 59, 128]) - layer.3.k_cache: torch.Size([1, 8, 59, 128]) - layer.3.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.k_cache: torch.Size([1, 8, 59, 128]) - layer.4.v_cache: torch.Size([1, 8, 59, 128]) - layer.4.output: torch.Size([1, 59, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02763205 104.83620233 - layer.0.v_cache 0.00000028 0.00070132 - layer.1.k_cache 0.00373147 7.45193740 - layer.1.v_cache 0.00000088 0.00279271 - layer.2.k_cache 0.00131257 2.40392084 - layer.2.v_cache 0.00000108 0.00421963 - layer.3.k_cache 0.00142120 2.86798354 - layer.3.v_cache 0.00000202 0.00684588 - layer.4.k_cache 0.00321287 7.03274278 - layer.4.v_cache 0.00000283 0.01062931 - layer.4.output 0.00027608 0.30931693 - ------------------------------------------------------------------------------------- - TOTAL 0.00274440 8.98966025 - (elements=845,824) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 845824 -Total Bytes 9436 -BPFP 0.0892 bits/point -EBPFP 0.1785 equivalent bits/point -MSE 8.989660 ----------------------- -------------------------------------------------------- -Time: 0.348s Load: 0.004s, Pack+Encode: 0.155s, Decode+Unpack: 0.190s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 59, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 59, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 8.9897 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample967-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample967-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample969-layer4-item1.zst (99/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample969-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 60, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.004s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 60, 128) -Output shape: (1, 60, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.0.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.1.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.1.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.2.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.2.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.3.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.3.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.4.k_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.4.v_cache: torch.Size([1, 8, 60, 128]) -> torch.Size([1, 1, 60, 1024]) - layer.4.output: torch.Size([1, 60, 4096]) -> torch.Size([1, 1, 60, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 568B, BPFP=0.0740 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 564B, BPFP=0.0734 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 724B, BPFP=0.0943 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 668B, BPFP=0.0870 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 756B, BPFP=0.0984 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 716B, BPFP=0.0932 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 776B, BPFP=0.1010 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 728B, BPFP=0.0948 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 712B, BPFP=0.0927 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 708B, BPFP=0.0922 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 2,252B, BPFP=0.0733 -⌛️ [2/4] FRONTEND: Frontend time: 0.147s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 60, 128]) - layer.0.v_cache: torch.Size([1, 8, 60, 128]) - layer.1.k_cache: torch.Size([1, 8, 60, 128]) - layer.1.v_cache: torch.Size([1, 8, 60, 128]) - layer.2.k_cache: torch.Size([1, 8, 60, 128]) - layer.2.v_cache: torch.Size([1, 8, 60, 128]) - layer.3.k_cache: torch.Size([1, 8, 60, 128]) - layer.3.v_cache: torch.Size([1, 8, 60, 128]) - layer.4.k_cache: torch.Size([1, 8, 60, 128]) - layer.4.v_cache: torch.Size([1, 8, 60, 128]) - layer.4.output: torch.Size([1, 60, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.190s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 60, 128]) - layer.0.v_cache: torch.Size([1, 8, 60, 128]) - layer.1.k_cache: torch.Size([1, 8, 60, 128]) - layer.1.v_cache: torch.Size([1, 8, 60, 128]) - layer.2.k_cache: torch.Size([1, 8, 60, 128]) - layer.2.v_cache: torch.Size([1, 8, 60, 128]) - layer.3.k_cache: torch.Size([1, 8, 60, 128]) - layer.3.v_cache: torch.Size([1, 8, 60, 128]) - layer.4.k_cache: torch.Size([1, 8, 60, 128]) - layer.4.v_cache: torch.Size([1, 8, 60, 128]) - layer.4.output: torch.Size([1, 60, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02844323 76.69359538 - layer.0.v_cache 0.00000027 0.00066808 - layer.1.k_cache 0.00360089 7.10871175 - layer.1.v_cache 0.00000087 0.00276349 - layer.2.k_cache 0.00116909 2.28062007 - layer.2.v_cache 0.00000119 0.00427924 - layer.3.k_cache 0.00139808 2.68885117 - layer.3.v_cache 0.00000212 0.00707959 - layer.4.k_cache 0.00324009 6.47276611 - layer.4.v_cache 0.00000310 0.01102089 - layer.4.output 0.00020032 0.27101517 - ------------------------------------------------------------------------------------- - TOTAL 0.00276144 6.88245832 - (elements=860,160) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 860160 -Total Bytes 9172 -BPFP 0.0853 bits/point -EBPFP 0.1706 equivalent bits/point -MSE 6.882458 ----------------------- -------------------------------------------------------- -Time: 0.342s Load: 0.004s, Pack+Encode: 0.147s, Decode+Unpack: 0.190s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 60, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 60, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.8825 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample969-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample969-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample97-layer4-item1.zst (100/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample97-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 114, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.007s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 114, 128) -Output shape: (1, 114, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.0.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.1.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.1.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.2.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.2.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.3.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.3.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.4.k_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.4.v_cache: torch.Size([1, 8, 114, 128]) -> torch.Size([1, 1, 114, 1024]) - layer.4.output: torch.Size([1, 114, 4096]) -> torch.Size([1, 1, 114, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,084B, BPFP=0.0743 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 1,104B, BPFP=0.0757 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 1,584B, BPFP=0.1086 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 1,332B, BPFP=0.0913 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 1,552B, BPFP=0.1064 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 1,524B, BPFP=0.1044 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 1,624B, BPFP=0.1113 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 1,520B, BPFP=0.1042 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 1,476B, BPFP=0.1012 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 1,528B, BPFP=0.1047 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 3,864B, BPFP=0.0662 -⌛️ [2/4] FRONTEND: Frontend time: 0.201s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 114, 128]) - layer.0.v_cache: torch.Size([1, 8, 114, 128]) - layer.1.k_cache: torch.Size([1, 8, 114, 128]) - layer.1.v_cache: torch.Size([1, 8, 114, 128]) - layer.2.k_cache: torch.Size([1, 8, 114, 128]) - layer.2.v_cache: torch.Size([1, 8, 114, 128]) - layer.3.k_cache: torch.Size([1, 8, 114, 128]) - layer.3.v_cache: torch.Size([1, 8, 114, 128]) - layer.4.k_cache: torch.Size([1, 8, 114, 128]) - layer.4.v_cache: torch.Size([1, 8, 114, 128]) - layer.4.output: torch.Size([1, 114, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.273s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 114, 128]) - layer.0.v_cache: torch.Size([1, 8, 114, 128]) - layer.1.k_cache: torch.Size([1, 8, 114, 128]) - layer.1.v_cache: torch.Size([1, 8, 114, 128]) - layer.2.k_cache: torch.Size([1, 8, 114, 128]) - layer.2.v_cache: torch.Size([1, 8, 114, 128]) - layer.3.k_cache: torch.Size([1, 8, 114, 128]) - layer.3.v_cache: torch.Size([1, 8, 114, 128]) - layer.4.k_cache: torch.Size([1, 8, 114, 128]) - layer.4.v_cache: torch.Size([1, 8, 114, 128]) - layer.4.output: torch.Size([1, 114, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435601 76.33458402 - layer.0.v_cache 0.00000028 0.00066030 - layer.1.k_cache 0.00337316 6.54298856 - layer.1.v_cache 0.00000076 0.00258455 - layer.2.k_cache 0.00114524 2.16346366 - layer.2.v_cache 0.00000110 0.00407046 - layer.3.k_cache 0.00139029 2.62863079 - layer.3.v_cache 0.00000222 0.00695160 - layer.4.k_cache 0.00339712 6.26660852 - layer.4.v_cache 0.00000302 0.01132016 - layer.4.output 0.00021100 0.24466183 - ------------------------------------------------------------------------------------- - TOTAL 0.00246523 6.78146500 - (elements=1,634,304) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 1634304 -Total Bytes 18192 -BPFP 0.0891 bits/point -EBPFP 0.1781 equivalent bits/point -MSE 6.781465 ----------------------- -------------------------------------------------------- -Time: 0.482s Load: 0.007s, Pack+Encode: 0.201s, Decode+Unpack: 0.273s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 114, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 114, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.7815 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_gsm8k/sample97-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_gsm8k/sample97-layer4-item1.zst ------------------------- ---------------------------- -TOTAL PROCESSING SUMMARY ------------------------- ---------------------------- -Total files 100 -Avg BPFP 0.1032 bits/point -Avg EBPFP 0.2063 equivalent bits/point -Avg MSE 9.209710 -Avg Time 0.499s ------------------------- ---------------------------- +version https://git-lfs.github.com/spec/v1 +oid sha256:7574dfeaa43501d74730b19a2a9835410fe054a154e44428b82a8ca10474b8a5 +size 1117768 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log b/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log index 3521854a0237b07529540763c295ca25f0ec3dab..39b073a95a160bb316da730c97623cafc0d4ec82 100644 --- a/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log @@ -1,16958 +1,3 @@ -Experiment: dtufc_hyperprior-featurecoding_qwen_individual -Log file: output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/dtufc_hyperprior-featurecoding_qwen_individual.log -DTUFCCodecConfig: - arch: hyperprior-featurecoding - handler: qwen - checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.001_epochs600_lr0.0001_bs360_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_lambda0.001_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar -Checkpoint epoch: 405 -Loaded hyperprior-featurecoding (1-channel) on cuda:0 -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.0.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_0_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.1.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_1_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.2.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_2_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.3.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_3_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.k_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_k.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.v_cache' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_layer_4_v.json -Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json: torch.Size([256]) -Loaded per-key quantization points for key 'layer.4.output' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/qwen3/arc_fewshot-8bit_feature.json -Loaded per-key mappings: model=qwen - Keys: ['layer.0.k_cache', 'layer.0.v_cache', 'layer.1.k_cache', 'layer.1.v_cache', 'layer.2.k_cache', 'layer.2.v_cache', 'layer.3.k_cache', 'layer.3.v_cache', 'layer.4.k_cache', 'layer.4.v_cache', 'layer.4.output'] ----------------- ------------------------------------------------------------------------------------------------------------------------------ -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding -Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.001_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar -Transform type kmeans -Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json -Input ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag -Output output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag ----------------- ------------------------------------------------------------------------------------------------------------------------------ -Files found: 100 ----------------------------------------------------------------------- - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample0-layer4-item1.zst (1/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample0-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 371, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.024s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 371, 128) -Output shape: (1, 371, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.0.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.1.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.1.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.2.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.2.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.3.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.3.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.4.k_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.4.v_cache: torch.Size([1, 8, 371, 128]) -> torch.Size([1, 1, 371, 1024]) - layer.4.output: torch.Size([1, 371, 4096]) -> torch.Size([1, 1, 371, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,752B, BPFP=0.0580 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,900B, BPFP=0.0611 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,900B, BPFP=0.0821 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,272B, BPFP=0.0689 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,084B, BPFP=0.0860 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,460B, BPFP=0.0729 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,276B, BPFP=0.0900 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,692B, BPFP=0.0777 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,780B, BPFP=0.0796 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,636B, BPFP=0.0766 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,904B, BPFP=0.0521 -⌛️ [2/4] FRONTEND: Frontend time: 0.841s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 371, 128]) - layer.0.v_cache: torch.Size([1, 8, 371, 128]) - layer.1.k_cache: torch.Size([1, 8, 371, 128]) - layer.1.v_cache: torch.Size([1, 8, 371, 128]) - layer.2.k_cache: torch.Size([1, 8, 371, 128]) - layer.2.v_cache: torch.Size([1, 8, 371, 128]) - layer.3.k_cache: torch.Size([1, 8, 371, 128]) - layer.3.v_cache: torch.Size([1, 8, 371, 128]) - layer.4.k_cache: torch.Size([1, 8, 371, 128]) - layer.4.v_cache: torch.Size([1, 8, 371, 128]) - layer.4.output: torch.Size([1, 371, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.712s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 371, 128]) - layer.0.v_cache: torch.Size([1, 8, 371, 128]) - layer.1.k_cache: torch.Size([1, 8, 371, 128]) - layer.1.v_cache: torch.Size([1, 8, 371, 128]) - layer.2.k_cache: torch.Size([1, 8, 371, 128]) - layer.2.v_cache: torch.Size([1, 8, 371, 128]) - layer.3.k_cache: torch.Size([1, 8, 371, 128]) - layer.3.v_cache: torch.Size([1, 8, 371, 128]) - layer.4.k_cache: torch.Size([1, 8, 371, 128]) - layer.4.v_cache: torch.Size([1, 8, 371, 128]) - layer.4.output: torch.Size([1, 371, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02382561 58.53507202 - layer.0.v_cache 0.00000026 0.00064548 - layer.1.k_cache 0.00289382 5.77842758 - layer.1.v_cache 0.00000075 0.00263513 - layer.2.k_cache 0.00115144 2.07946909 - layer.2.v_cache 0.00000114 0.00415350 - layer.3.k_cache 0.00133317 2.43022053 - layer.3.v_cache 0.00000213 0.00707252 - layer.4.k_cache 0.00354181 6.38041584 - layer.4.v_cache 0.00000324 0.01163471 - layer.4.output 0.00015639 0.19191734 - ------------------------------------------------------------------------------------- - TOTAL 0.00238421 5.42838684 - (elements=5,318,656) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5318656 -Total Bytes 45656 -BPFP 0.0687 bits/point -EBPFP 0.1373 equivalent bits/point -MSE 5.428387 ----------------------- -------------------------------------------------------- -Time: 1.578s Load: 0.024s, Pack+Encode: 0.841s, Decode+Unpack: 0.712s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 371, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 371, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.4284 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample0-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample0-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample1-layer4-item1.zst (2/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample1-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,592B, BPFP=0.0589 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,844B, BPFP=0.0646 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,744B, BPFP=0.0850 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,292B, BPFP=0.0748 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,020B, BPFP=0.0913 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,440B, BPFP=0.0781 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,140B, BPFP=0.0940 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,640B, BPFP=0.0827 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,660B, BPFP=0.0831 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,560B, BPFP=0.0809 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,388B, BPFP=0.0590 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.672s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02409526 55.32891420 - layer.0.v_cache 0.00000026 0.00065984 - layer.1.k_cache 0.00287302 5.70794536 - layer.1.v_cache 0.00000078 0.00276465 - layer.2.k_cache 0.00123767 2.12195694 - layer.2.v_cache 0.00000117 0.00445075 - layer.3.k_cache 0.00130179 2.44654669 - layer.3.v_cache 0.00000226 0.00755731 - layer.4.k_cache 0.00355144 6.38725387 - layer.4.v_cache 0.00000325 0.01219581 - layer.4.output 0.00015063 0.19821189 - ------------------------------------------------------------------------------------- - TOTAL 0.00240496 5.20093521 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 45320 -BPFP 0.0735 bits/point -EBPFP 0.1470 equivalent bits/point -MSE 5.200935 ----------------------- -------------------------------------------------------- -Time: 1.096s Load: 0.018s, Pack+Encode: 0.406s, Decode+Unpack: 0.672s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.2009 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample1-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample1-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample10-layer4-item1.zst (3/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample10-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 377, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 377, 128) -Output shape: (1, 377, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.0.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.1.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.1.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.2.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.2.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.3.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.3.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.4.k_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.4.v_cache: torch.Size([1, 8, 377, 128]) -> torch.Size([1, 1, 377, 1024]) - layer.4.output: torch.Size([1, 377, 4096]) -> torch.Size([1, 1, 377, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,656B, BPFP=0.0550 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,800B, BPFP=0.0580 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,764B, BPFP=0.0780 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,264B, BPFP=0.0676 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,032B, BPFP=0.0836 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,460B, BPFP=0.0717 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,188B, BPFP=0.0868 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,684B, BPFP=0.0763 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,660B, BPFP=0.0758 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,628B, BPFP=0.0752 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,780B, BPFP=0.0507 -⌛️ [2/4] FRONTEND: Frontend time: 0.410s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 377, 128]) - layer.0.v_cache: torch.Size([1, 8, 377, 128]) - layer.1.k_cache: torch.Size([1, 8, 377, 128]) - layer.1.v_cache: torch.Size([1, 8, 377, 128]) - layer.2.k_cache: torch.Size([1, 8, 377, 128]) - layer.2.v_cache: torch.Size([1, 8, 377, 128]) - layer.3.k_cache: torch.Size([1, 8, 377, 128]) - layer.3.v_cache: torch.Size([1, 8, 377, 128]) - layer.4.k_cache: torch.Size([1, 8, 377, 128]) - layer.4.v_cache: torch.Size([1, 8, 377, 128]) - layer.4.output: torch.Size([1, 377, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.684s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 377, 128]) - layer.0.v_cache: torch.Size([1, 8, 377, 128]) - layer.1.k_cache: torch.Size([1, 8, 377, 128]) - layer.1.v_cache: torch.Size([1, 8, 377, 128]) - layer.2.k_cache: torch.Size([1, 8, 377, 128]) - layer.2.v_cache: torch.Size([1, 8, 377, 128]) - layer.3.k_cache: torch.Size([1, 8, 377, 128]) - layer.3.v_cache: torch.Size([1, 8, 377, 128]) - layer.4.k_cache: torch.Size([1, 8, 377, 128]) - layer.4.v_cache: torch.Size([1, 8, 377, 128]) - layer.4.output: torch.Size([1, 377, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02400962 76.54609271 - layer.0.v_cache 0.00000027 0.00068551 - layer.1.k_cache 0.00291640 6.20941308 - layer.1.v_cache 0.00000080 0.00288822 - layer.2.k_cache 0.00116459 2.12857485 - layer.2.v_cache 0.00000114 0.00443793 - layer.3.k_cache 0.00131780 2.46214978 - layer.3.v_cache 0.00000210 0.00736165 - layer.4.k_cache 0.00355012 6.72165896 - layer.4.v_cache 0.00000332 0.01264759 - layer.4.output 0.00014433 0.19624531 - ------------------------------------------------------------------------------------- - TOTAL 0.00239596 6.77720654 - (elements=5,404,672) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5404672 -Total Bytes 44916 -BPFP 0.0665 bits/point -EBPFP 0.1330 equivalent bits/point -MSE 6.777207 ----------------------- -------------------------------------------------------- -Time: 1.113s Load: 0.019s, Pack+Encode: 0.410s, Decode+Unpack: 0.684s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 377, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 377, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.7772 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample10-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample10-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample100-layer4-item1.zst (4/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample100-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 315, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.015s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 315, 128) -Output shape: (1, 315, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.0.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.1.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.1.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.2.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.2.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.3.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.3.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.4.k_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.4.v_cache: torch.Size([1, 8, 315, 128]) -> torch.Size([1, 1, 315, 1024]) - layer.4.output: torch.Size([1, 315, 4096]) -> torch.Size([1, 1, 315, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,224B, BPFP=0.0552 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,340B, BPFP=0.0580 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,088B, BPFP=0.0766 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,700B, BPFP=0.0670 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,304B, BPFP=0.0819 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,812B, BPFP=0.0697 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,416B, BPFP=0.0847 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,996B, BPFP=0.0743 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,984B, BPFP=0.0740 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,912B, BPFP=0.0722 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 8,304B, BPFP=0.0515 -⌛️ [2/4] FRONTEND: Frontend time: 0.467s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 315, 128]) - layer.0.v_cache: torch.Size([1, 8, 315, 128]) - layer.1.k_cache: torch.Size([1, 8, 315, 128]) - layer.1.v_cache: torch.Size([1, 8, 315, 128]) - layer.2.k_cache: torch.Size([1, 8, 315, 128]) - layer.2.v_cache: torch.Size([1, 8, 315, 128]) - layer.3.k_cache: torch.Size([1, 8, 315, 128]) - layer.3.v_cache: torch.Size([1, 8, 315, 128]) - layer.4.k_cache: torch.Size([1, 8, 315, 128]) - layer.4.v_cache: torch.Size([1, 8, 315, 128]) - layer.4.output: torch.Size([1, 315, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.577s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 315, 128]) - layer.0.v_cache: torch.Size([1, 8, 315, 128]) - layer.1.k_cache: torch.Size([1, 8, 315, 128]) - layer.1.v_cache: torch.Size([1, 8, 315, 128]) - layer.2.k_cache: torch.Size([1, 8, 315, 128]) - layer.2.v_cache: torch.Size([1, 8, 315, 128]) - layer.3.k_cache: torch.Size([1, 8, 315, 128]) - layer.3.v_cache: torch.Size([1, 8, 315, 128]) - layer.4.k_cache: torch.Size([1, 8, 315, 128]) - layer.4.v_cache: torch.Size([1, 8, 315, 128]) - layer.4.output: torch.Size([1, 315, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02468849 49.45152220 - layer.0.v_cache 0.00000027 0.00067431 - layer.1.k_cache 0.00284817 5.72493218 - layer.1.v_cache 0.00000082 0.00282307 - layer.2.k_cache 0.00122538 2.10377934 - layer.2.v_cache 0.00000116 0.00436578 - layer.3.k_cache 0.00131111 2.43964689 - layer.3.v_cache 0.00000216 0.00744839 - layer.4.k_cache 0.00356899 6.42766733 - layer.4.v_cache 0.00000330 0.01231051 - layer.4.output 0.00013845 0.19346040 - ------------------------------------------------------------------------------------- - TOTAL 0.00244312 4.78207226 - (elements=4,515,840) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4515840 -Total Bytes 37080 -BPFP 0.0657 bits/point -EBPFP 0.1314 equivalent bits/point -MSE 4.782072 ----------------------- -------------------------------------------------------- -Time: 1.059s Load: 0.015s, Pack+Encode: 0.467s, Decode+Unpack: 0.577s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 315, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 315, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7821 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample100-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample100-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample102-layer4-item1.zst (5/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample102-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 362, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 362, 128) -Output shape: (1, 362, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.output: torch.Size([1, 362, 4096]) -> torch.Size([1, 1, 362, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,760B, BPFP=0.0596 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,856B, BPFP=0.0616 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,900B, BPFP=0.0842 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,332B, BPFP=0.0719 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,124B, BPFP=0.0890 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,500B, BPFP=0.0755 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,304B, BPFP=0.0929 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,748B, BPFP=0.0809 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,752B, BPFP=0.0810 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,628B, BPFP=0.0783 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,940B, BPFP=0.0536 -⌛️ [2/4] FRONTEND: Frontend time: 0.408s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.output: torch.Size([1, 362, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.673s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.output: torch.Size([1, 362, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02398593 55.96729325 - layer.0.v_cache 0.00000027 0.00066626 - layer.1.k_cache 0.00294127 5.90327243 - layer.1.v_cache 0.00000079 0.00280733 - layer.2.k_cache 0.00116280 2.12401265 - layer.2.v_cache 0.00000114 0.00434748 - layer.3.k_cache 0.00131002 2.43557925 - layer.3.v_cache 0.00000215 0.00732663 - layer.4.k_cache 0.00366934 6.47846976 - layer.4.v_cache 0.00000328 0.01221378 - layer.4.output 0.00013411 0.18257611 - ------------------------------------------------------------------------------------- - TOTAL 0.00240096 5.26187809 - (elements=5,189,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5189632 -Total Bytes 45844 -BPFP 0.0707 bits/point -EBPFP 0.1413 equivalent bits/point -MSE 5.261878 ----------------------- -------------------------------------------------------- -Time: 1.098s Load: 0.017s, Pack+Encode: 0.408s, Decode+Unpack: 0.673s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 362, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.2619 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample102-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample102-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample105-layer4-item1.zst (6/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample105-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 336, 128) -Output shape: (1, 336, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.output: torch.Size([1, 336, 4096]) -> torch.Size([1, 1, 336, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,552B, BPFP=0.0593 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,764B, BPFP=0.0643 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,796B, BPFP=0.0883 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,184B, BPFP=0.0740 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,012B, BPFP=0.0933 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,400B, BPFP=0.0791 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,200B, BPFP=0.0977 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,680B, BPFP=0.0856 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,592B, BPFP=0.0835 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,536B, BPFP=0.0822 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,152B, BPFP=0.0590 -⌛️ [2/4] FRONTEND: Frontend time: 0.407s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.672s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02466884 48.70736840 - layer.0.v_cache 0.00000026 0.00065547 - layer.1.k_cache 0.00293366 5.85656993 - layer.1.v_cache 0.00000078 0.00271375 - layer.2.k_cache 0.00117059 2.09419923 - layer.2.v_cache 0.00000113 0.00424644 - layer.3.k_cache 0.00132874 2.43271147 - layer.3.v_cache 0.00000215 0.00732780 - layer.4.k_cache 0.00360252 6.44871085 - layer.4.v_cache 0.00000314 0.01163969 - layer.4.output 0.00014135 0.19627835 - ------------------------------------------------------------------------------------- - TOTAL 0.00244837 4.73937546 - (elements=4,816,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4816896 -Total Bytes 44868 -BPFP 0.0745 bits/point -EBPFP 0.1490 equivalent bits/point -MSE 4.739375 ----------------------- -------------------------------------------------------- -Time: 1.095s Load: 0.017s, Pack+Encode: 0.407s, Decode+Unpack: 0.672s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7394 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample105-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample105-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample108-layer4-item1.zst (7/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample108-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 317, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 317, 128) -Output shape: (1, 317, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.0.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.1.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.1.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.2.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.2.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.3.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.3.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.4.k_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.4.v_cache: torch.Size([1, 8, 317, 128]) -> torch.Size([1, 1, 317, 1024]) - layer.4.output: torch.Size([1, 317, 4096]) -> torch.Size([1, 1, 317, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,112B, BPFP=0.0521 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,168B, BPFP=0.0534 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,912B, BPFP=0.0718 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,496B, BPFP=0.0615 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,148B, BPFP=0.0776 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,608B, BPFP=0.0643 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,256B, BPFP=0.0802 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,800B, BPFP=0.0690 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,808B, BPFP=0.0692 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,728B, BPFP=0.0672 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 7,744B, BPFP=0.0477 -⌛️ [2/4] FRONTEND: Frontend time: 0.356s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 317, 128]) - layer.0.v_cache: torch.Size([1, 8, 317, 128]) - layer.1.k_cache: torch.Size([1, 8, 317, 128]) - layer.1.v_cache: torch.Size([1, 8, 317, 128]) - layer.2.k_cache: torch.Size([1, 8, 317, 128]) - layer.2.v_cache: torch.Size([1, 8, 317, 128]) - layer.3.k_cache: torch.Size([1, 8, 317, 128]) - layer.3.v_cache: torch.Size([1, 8, 317, 128]) - layer.4.k_cache: torch.Size([1, 8, 317, 128]) - layer.4.v_cache: torch.Size([1, 8, 317, 128]) - layer.4.output: torch.Size([1, 317, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.576s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 317, 128]) - layer.0.v_cache: torch.Size([1, 8, 317, 128]) - layer.1.k_cache: torch.Size([1, 8, 317, 128]) - layer.1.v_cache: torch.Size([1, 8, 317, 128]) - layer.2.k_cache: torch.Size([1, 8, 317, 128]) - layer.2.v_cache: torch.Size([1, 8, 317, 128]) - layer.3.k_cache: torch.Size([1, 8, 317, 128]) - layer.3.v_cache: torch.Size([1, 8, 317, 128]) - layer.4.k_cache: torch.Size([1, 8, 317, 128]) - layer.4.v_cache: torch.Size([1, 8, 317, 128]) - layer.4.output: torch.Size([1, 317, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02499472 42.22667340 - layer.0.v_cache 0.00000027 0.00065328 - layer.1.k_cache 0.00294942 5.77249829 - layer.1.v_cache 0.00000081 0.00279394 - layer.2.k_cache 0.00116222 2.05313120 - layer.2.v_cache 0.00000115 0.00435366 - layer.3.k_cache 0.00131165 2.40130923 - layer.3.v_cache 0.00000215 0.00736699 - layer.4.k_cache 0.00349495 6.23641900 - layer.4.v_cache 0.00000332 0.01217810 - layer.4.output 0.00014092 0.19284308 - ------------------------------------------------------------------------------------- - TOTAL 0.00246317 4.24919639 - (elements=4,544,512) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4544512 -Total Bytes 34780 -BPFP 0.0612 bits/point -EBPFP 0.1225 equivalent bits/point -MSE 4.249196 ----------------------- -------------------------------------------------------- -Time: 0.949s Load: 0.017s, Pack+Encode: 0.356s, Decode+Unpack: 0.576s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 317, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 317, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.2492 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample108-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample108-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample11-layer4-item1.zst (8/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample11-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 353, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 353, 128) -Output shape: (1, 353, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.output: torch.Size([1, 353, 4096]) -> torch.Size([1, 1, 353, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,504B, BPFP=0.0554 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,640B, BPFP=0.0584 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,644B, BPFP=0.0806 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,092B, BPFP=0.0684 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,904B, BPFP=0.0864 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,332B, BPFP=0.0737 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,064B, BPFP=0.0899 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,524B, BPFP=0.0780 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,532B, BPFP=0.0782 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,380B, BPFP=0.0748 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,536B, BPFP=0.0528 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.output: torch.Size([1, 353, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.669s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.output: torch.Size([1, 353, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02498756 61.32225456 - layer.0.v_cache 0.00000027 0.00066741 - layer.1.k_cache 0.00288490 5.79649565 - layer.1.v_cache 0.00000079 0.00282481 - layer.2.k_cache 0.00118233 2.12205838 - layer.2.v_cache 0.00000115 0.00438574 - layer.3.k_cache 0.00132883 2.45056809 - layer.3.v_cache 0.00000217 0.00717708 - layer.4.k_cache 0.00368366 6.57672517 - layer.4.v_cache 0.00000317 0.01189352 - layer.4.output 0.00013605 0.19090094 - ------------------------------------------------------------------------------------- - TOTAL 0.00247279 5.64704673 - (elements=5,060,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5060608 -Total Bytes 43152 -BPFP 0.0682 bits/point -EBPFP 0.1364 equivalent bits/point -MSE 5.647047 ----------------------- -------------------------------------------------------- -Time: 1.093s Load: 0.018s, Pack+Encode: 0.406s, Decode+Unpack: 0.669s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 353, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.6470 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample11-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample11-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample112-layer4-item1.zst (9/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample112-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 318, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 318, 128) -Output shape: (1, 318, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.output: torch.Size([1, 318, 4096]) -> torch.Size([1, 1, 318, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,028B, BPFP=0.0498 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,128B, BPFP=0.0523 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,856B, BPFP=0.0702 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,340B, BPFP=0.0575 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,116B, BPFP=0.0766 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,528B, BPFP=0.0621 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,228B, BPFP=0.0793 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,700B, BPFP=0.0663 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,752B, BPFP=0.0676 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,640B, BPFP=0.0649 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 8,208B, BPFP=0.0504 -⌛️ [2/4] FRONTEND: Frontend time: 0.356s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.output: torch.Size([1, 318, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.569s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.output: torch.Size([1, 318, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02461332 38.08535402 - layer.0.v_cache 0.00000025 0.00063750 - layer.1.k_cache 0.00290030 5.60263426 - layer.1.v_cache 0.00000077 0.00249154 - layer.2.k_cache 0.00116467 2.02441733 - layer.2.v_cache 0.00000108 0.00392365 - layer.3.k_cache 0.00132360 2.34736912 - layer.3.v_cache 0.00000220 0.00691997 - layer.4.k_cache 0.00367546 6.31705460 - layer.4.v_cache 0.00000314 0.01149241 - layer.4.output 0.00019099 0.21630564 - ------------------------------------------------------------------------------------- - TOTAL 0.00246062 3.94767978 - (elements=4,558,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4558848 -Total Bytes 34524 -BPFP 0.0606 bits/point -EBPFP 0.1212 equivalent bits/point -MSE 3.947680 ----------------------- -------------------------------------------------------- -Time: 0.940s Load: 0.016s, Pack+Encode: 0.356s, Decode+Unpack: 0.569s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 318, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 3.9477 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample112-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample112-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample114-layer4-item1.zst (10/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample114-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,624B, BPFP=0.0621 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,804B, BPFP=0.0664 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,768B, BPFP=0.0892 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,288B, BPFP=0.0778 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,096B, BPFP=0.0970 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,564B, BPFP=0.0844 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,200B, BPFP=0.0994 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,768B, BPFP=0.0892 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,632B, BPFP=0.0860 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,624B, BPFP=0.0858 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,460B, BPFP=0.0619 -⌛️ [2/4] FRONTEND: Frontend time: 0.404s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.669s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02377391 49.56689157 - layer.0.v_cache 0.00000027 0.00067908 - layer.1.k_cache 0.00289258 5.94671927 - layer.1.v_cache 0.00000083 0.00280466 - layer.2.k_cache 0.00117497 2.11560632 - layer.2.v_cache 0.00000126 0.00446728 - layer.3.k_cache 0.00131231 2.45093162 - layer.3.v_cache 0.00000230 0.00759633 - layer.4.k_cache 0.00357793 6.51070816 - layer.4.v_cache 0.00000347 0.01250676 - layer.4.output 0.00014070 0.19560598 - ------------------------------------------------------------------------------------- - TOTAL 0.00237876 4.81438107 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 45828 -BPFP 0.0775 bits/point -EBPFP 0.1550 equivalent bits/point -MSE 4.814381 ----------------------- -------------------------------------------------------- -Time: 1.090s Load: 0.017s, Pack+Encode: 0.404s, Decode+Unpack: 0.669s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8144 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample114-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample114-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample119-layer4-item1.zst (11/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample119-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 357, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 357, 128) -Output shape: (1, 357, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.0.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.1.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.1.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.2.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.2.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.3.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.3.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.4.k_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.4.v_cache: torch.Size([1, 8, 357, 128]) -> torch.Size([1, 1, 357, 1024]) - layer.4.output: torch.Size([1, 357, 4096]) -> torch.Size([1, 1, 357, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,616B, BPFP=0.0572 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,824B, BPFP=0.0618 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,768B, BPFP=0.0825 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,240B, BPFP=0.0709 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,952B, BPFP=0.0865 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,392B, BPFP=0.0742 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,136B, BPFP=0.0905 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,652B, BPFP=0.0799 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,636B, BPFP=0.0796 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,544B, BPFP=0.0776 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,412B, BPFP=0.0570 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 357, 128]) - layer.0.v_cache: torch.Size([1, 8, 357, 128]) - layer.1.k_cache: torch.Size([1, 8, 357, 128]) - layer.1.v_cache: torch.Size([1, 8, 357, 128]) - layer.2.k_cache: torch.Size([1, 8, 357, 128]) - layer.2.v_cache: torch.Size([1, 8, 357, 128]) - layer.3.k_cache: torch.Size([1, 8, 357, 128]) - layer.3.v_cache: torch.Size([1, 8, 357, 128]) - layer.4.k_cache: torch.Size([1, 8, 357, 128]) - layer.4.v_cache: torch.Size([1, 8, 357, 128]) - layer.4.output: torch.Size([1, 357, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.662s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 357, 128]) - layer.0.v_cache: torch.Size([1, 8, 357, 128]) - layer.1.k_cache: torch.Size([1, 8, 357, 128]) - layer.1.v_cache: torch.Size([1, 8, 357, 128]) - layer.2.k_cache: torch.Size([1, 8, 357, 128]) - layer.2.v_cache: torch.Size([1, 8, 357, 128]) - layer.3.k_cache: torch.Size([1, 8, 357, 128]) - layer.3.v_cache: torch.Size([1, 8, 357, 128]) - layer.4.k_cache: torch.Size([1, 8, 357, 128]) - layer.4.v_cache: torch.Size([1, 8, 357, 128]) - layer.4.output: torch.Size([1, 357, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02468803 48.80272343 - layer.0.v_cache 0.00000027 0.00068724 - layer.1.k_cache 0.00290374 6.02150694 - layer.1.v_cache 0.00000082 0.00276899 - layer.2.k_cache 0.00112985 2.05788578 - layer.2.v_cache 0.00000115 0.00418538 - layer.3.k_cache 0.00131122 2.45700834 - layer.3.v_cache 0.00000220 0.00740602 - layer.4.k_cache 0.00345488 6.21847714 - layer.4.v_cache 0.00000325 0.01210617 - layer.4.output 0.00016267 0.20809112 - ------------------------------------------------------------------------------------- - TOTAL 0.00243901 4.74407999 - (elements=5,117,952) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5117952 -Total Bytes 45172 -BPFP 0.0706 bits/point -EBPFP 0.1412 equivalent bits/point -MSE 4.744080 ----------------------- -------------------------------------------------------- -Time: 1.086s Load: 0.017s, Pack+Encode: 0.406s, Decode+Unpack: 0.662s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 357, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 357, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7441 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample119-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample119-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample12-layer4-item1.zst (12/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample12-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 358, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 358, 128) -Output shape: (1, 358, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.0.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.1.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.1.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.2.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.2.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.3.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.3.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.4.k_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.4.v_cache: torch.Size([1, 8, 358, 128]) -> torch.Size([1, 1, 358, 1024]) - layer.4.output: torch.Size([1, 358, 4096]) -> torch.Size([1, 1, 358, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,640B, BPFP=0.0576 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,840B, BPFP=0.0620 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,800B, BPFP=0.0829 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,300B, BPFP=0.0720 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,000B, BPFP=0.0873 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,504B, BPFP=0.0765 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,208B, BPFP=0.0918 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,708B, BPFP=0.0809 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,672B, BPFP=0.0801 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,608B, BPFP=0.0787 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,168B, BPFP=0.0555 -⌛️ [2/4] FRONTEND: Frontend time: 0.403s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 358, 128]) - layer.0.v_cache: torch.Size([1, 8, 358, 128]) - layer.1.k_cache: torch.Size([1, 8, 358, 128]) - layer.1.v_cache: torch.Size([1, 8, 358, 128]) - layer.2.k_cache: torch.Size([1, 8, 358, 128]) - layer.2.v_cache: torch.Size([1, 8, 358, 128]) - layer.3.k_cache: torch.Size([1, 8, 358, 128]) - layer.3.v_cache: torch.Size([1, 8, 358, 128]) - layer.4.k_cache: torch.Size([1, 8, 358, 128]) - layer.4.v_cache: torch.Size([1, 8, 358, 128]) - layer.4.output: torch.Size([1, 358, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.661s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 358, 128]) - layer.0.v_cache: torch.Size([1, 8, 358, 128]) - layer.1.k_cache: torch.Size([1, 8, 358, 128]) - layer.1.v_cache: torch.Size([1, 8, 358, 128]) - layer.2.k_cache: torch.Size([1, 8, 358, 128]) - layer.2.v_cache: torch.Size([1, 8, 358, 128]) - layer.3.k_cache: torch.Size([1, 8, 358, 128]) - layer.3.v_cache: torch.Size([1, 8, 358, 128]) - layer.4.k_cache: torch.Size([1, 8, 358, 128]) - layer.4.v_cache: torch.Size([1, 8, 358, 128]) - layer.4.output: torch.Size([1, 358, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02418174 50.56650445 - layer.0.v_cache 0.00000026 0.00066488 - layer.1.k_cache 0.00288075 5.76629332 - layer.1.v_cache 0.00000077 0.00272360 - layer.2.k_cache 0.00118336 2.07629122 - layer.2.v_cache 0.00000117 0.00432719 - layer.3.k_cache 0.00130988 2.41011925 - layer.3.v_cache 0.00000211 0.00726196 - layer.4.k_cache 0.00353008 6.47433386 - layer.4.v_cache 0.00000317 0.01211180 - layer.4.output 0.00013834 0.19028000 - ------------------------------------------------------------------------------------- - TOTAL 0.00240333 4.86298225 - (elements=5,132,288) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5132288 -Total Bytes 45448 -BPFP 0.0708 bits/point -EBPFP 0.1417 equivalent bits/point -MSE 4.862982 ----------------------- -------------------------------------------------------- -Time: 1.083s Load: 0.019s, Pack+Encode: 0.403s, Decode+Unpack: 0.661s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 358, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 358, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8630 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample12-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample12-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample13-layer4-item1.zst (13/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample13-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 336, 128) -Output shape: (1, 336, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.output: torch.Size([1, 336, 4096]) -> torch.Size([1, 1, 336, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,580B, BPFP=0.0600 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,840B, BPFP=0.0660 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,824B, BPFP=0.0889 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,340B, BPFP=0.0777 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,048B, BPFP=0.0941 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,516B, BPFP=0.0818 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,212B, BPFP=0.0979 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,732B, BPFP=0.0868 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,628B, BPFP=0.0844 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,648B, BPFP=0.0848 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,732B, BPFP=0.0624 -⌛️ [2/4] FRONTEND: Frontend time: 0.403s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.665s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02407829 47.18049840 - layer.0.v_cache 0.00000026 0.00064415 - layer.1.k_cache 0.00285290 5.87168521 - layer.1.v_cache 0.00000077 0.00277587 - layer.2.k_cache 0.00116461 2.11610231 - layer.2.v_cache 0.00000115 0.00439830 - layer.3.k_cache 0.00131253 2.44226220 - layer.3.v_cache 0.00000220 0.00745028 - layer.4.k_cache 0.00369396 6.57188270 - layer.4.v_cache 0.00000328 0.01223648 - layer.4.output 0.00015589 0.20660407 - ------------------------------------------------------------------------------------- - TOTAL 0.00240954 4.64545373 - (elements=4,816,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4816896 -Total Bytes 46100 -BPFP 0.0766 bits/point -EBPFP 0.1531 equivalent bits/point -MSE 4.645454 ----------------------- -------------------------------------------------------- -Time: 1.084s Load: 0.016s, Pack+Encode: 0.403s, Decode+Unpack: 0.665s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.6455 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample13-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample13-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample132-layer4-item1.zst (14/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample132-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 346, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 346, 128) -Output shape: (1, 346, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.output: torch.Size([1, 346, 4096]) -> torch.Size([1, 1, 346, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,532B, BPFP=0.0572 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,720B, BPFP=0.0614 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,680B, BPFP=0.0831 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,140B, BPFP=0.0709 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,952B, BPFP=0.0892 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,352B, BPFP=0.0757 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,080B, BPFP=0.0921 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,540B, BPFP=0.0799 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,536B, BPFP=0.0798 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,472B, BPFP=0.0784 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,540B, BPFP=0.0595 -⌛️ [2/4] FRONTEND: Frontend time: 0.425s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.output: torch.Size([1, 346, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.677s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.output: torch.Size([1, 346, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02507911 62.37271947 - layer.0.v_cache 0.00000027 0.00067256 - layer.1.k_cache 0.00290669 5.92871799 - layer.1.v_cache 0.00000082 0.00276197 - layer.2.k_cache 0.00115107 2.05428918 - layer.2.v_cache 0.00000117 0.00427462 - layer.3.k_cache 0.00130922 2.42831456 - layer.3.v_cache 0.00000214 0.00731433 - layer.4.k_cache 0.00347357 6.26985089 - layer.4.v_cache 0.00000325 0.01199288 - layer.4.output 0.00015455 0.20878160 - ------------------------------------------------------------------------------------- - TOTAL 0.00246754 5.70828820 - (elements=4,960,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4960256 -Total Bytes 44544 -BPFP 0.0718 bits/point -EBPFP 0.1437 equivalent bits/point -MSE 5.708288 ----------------------- -------------------------------------------------------- -Time: 1.120s Load: 0.018s, Pack+Encode: 0.425s, Decode+Unpack: 0.677s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 346, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.7083 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample132-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample132-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample135-layer4-item1.zst (15/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample135-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 343, 128) -Output shape: (1, 343, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.output: torch.Size([1, 343, 4096]) -> torch.Size([1, 1, 343, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,596B, BPFP=0.0591 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,748B, BPFP=0.0626 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,740B, BPFP=0.0852 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,172B, BPFP=0.0722 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,008B, BPFP=0.0913 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,372B, BPFP=0.0768 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,148B, BPFP=0.0945 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,604B, BPFP=0.0821 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,624B, BPFP=0.0825 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,508B, BPFP=0.0799 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,972B, BPFP=0.0568 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.664s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02309080 61.07775487 - layer.0.v_cache 0.00000026 0.00064572 - layer.1.k_cache 0.00292563 6.00812426 - layer.1.v_cache 0.00000076 0.00272505 - layer.2.k_cache 0.00117950 2.09374751 - layer.2.v_cache 0.00000116 0.00417370 - layer.3.k_cache 0.00131686 2.42438011 - layer.3.v_cache 0.00000208 0.00705998 - layer.4.k_cache 0.00357410 6.48553591 - layer.4.v_cache 0.00000334 0.01173045 - layer.4.output 0.00016206 0.19513313 - ------------------------------------------------------------------------------------- - TOTAL 0.00233876 5.63545786 - (elements=4,917,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4917248 -Total Bytes 44492 -BPFP 0.0724 bits/point -EBPFP 0.1448 equivalent bits/point -MSE 5.635458 ----------------------- -------------------------------------------------------- -Time: 1.088s Load: 0.017s, Pack+Encode: 0.406s, Decode+Unpack: 0.664s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.6355 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample135-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample135-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample14-layer4-item1.zst (16/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample14-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 359, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 359, 128) -Output shape: (1, 359, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.output: torch.Size([1, 359, 4096]) -> torch.Size([1, 1, 359, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,692B, BPFP=0.0586 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,896B, BPFP=0.0630 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,860B, BPFP=0.0840 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,380B, BPFP=0.0736 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,056B, BPFP=0.0883 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,564B, BPFP=0.0776 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,236B, BPFP=0.0922 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,764B, BPFP=0.0819 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,716B, BPFP=0.0809 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,640B, BPFP=0.0792 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,940B, BPFP=0.0541 -⌛️ [2/4] FRONTEND: Frontend time: 0.407s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.output: torch.Size([1, 359, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.670s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.output: torch.Size([1, 359, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02394788 49.96947902 - layer.0.v_cache 0.00000027 0.00066129 - layer.1.k_cache 0.00287336 5.91911533 - layer.1.v_cache 0.00000081 0.00292834 - layer.2.k_cache 0.00118844 2.12506036 - layer.2.v_cache 0.00000120 0.00451918 - layer.3.k_cache 0.00133454 2.44597641 - layer.3.v_cache 0.00000220 0.00750061 - layer.4.k_cache 0.00356152 6.45646340 - layer.4.v_cache 0.00000363 0.01249352 - layer.4.output 0.00015158 0.18950285 - ------------------------------------------------------------------------------------- - TOTAL 0.00239430 4.83587206 - (elements=5,146,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5146624 -Total Bytes 45744 -BPFP 0.0711 bits/point -EBPFP 0.1422 equivalent bits/point -MSE 4.835872 ----------------------- -------------------------------------------------------- -Time: 1.095s Load: 0.018s, Pack+Encode: 0.407s, Decode+Unpack: 0.670s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 359, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8359 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample14-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample14-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample15-layer4-item1.zst (17/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample15-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 359, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 359, 128) -Output shape: (1, 359, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) -> torch.Size([1, 1, 359, 1024]) - layer.4.output: torch.Size([1, 359, 4096]) -> torch.Size([1, 1, 359, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,676B, BPFP=0.0582 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,872B, BPFP=0.0625 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,872B, BPFP=0.0843 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,444B, BPFP=0.0749 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,056B, BPFP=0.0883 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,576B, BPFP=0.0778 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,244B, BPFP=0.0924 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,800B, BPFP=0.0827 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,720B, BPFP=0.0810 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,672B, BPFP=0.0799 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,344B, BPFP=0.0563 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.output: torch.Size([1, 359, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.658s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 359, 128]) - layer.0.v_cache: torch.Size([1, 8, 359, 128]) - layer.1.k_cache: torch.Size([1, 8, 359, 128]) - layer.1.v_cache: torch.Size([1, 8, 359, 128]) - layer.2.k_cache: torch.Size([1, 8, 359, 128]) - layer.2.v_cache: torch.Size([1, 8, 359, 128]) - layer.3.k_cache: torch.Size([1, 8, 359, 128]) - layer.3.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.k_cache: torch.Size([1, 8, 359, 128]) - layer.4.v_cache: torch.Size([1, 8, 359, 128]) - layer.4.output: torch.Size([1, 359, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02338462 52.66365447 - layer.0.v_cache 0.00000027 0.00066932 - layer.1.k_cache 0.00293092 5.99203449 - layer.1.v_cache 0.00000080 0.00281893 - layer.2.k_cache 0.00116855 2.10543373 - layer.2.v_cache 0.00000117 0.00438799 - layer.3.k_cache 0.00131991 2.43780127 - layer.3.v_cache 0.00000231 0.00753909 - layer.4.k_cache 0.00358756 6.40399901 - layer.4.v_cache 0.00000356 0.01246874 - layer.4.output 0.00016308 0.19046080 - ------------------------------------------------------------------------------------- - TOTAL 0.00236086 5.02804644 - (elements=5,146,624) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5146624 -Total Bytes 46276 -BPFP 0.0719 bits/point -EBPFP 0.1439 equivalent bits/point -MSE 5.028046 ----------------------- -------------------------------------------------------- -Time: 1.092s Load: 0.019s, Pack+Encode: 0.414s, Decode+Unpack: 0.658s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 359, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 359, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.0280 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample15-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample15-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample16-layer4-item1.zst (18/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample16-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 347, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 347, 128) -Output shape: (1, 347, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.0.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.1.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.1.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.2.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.2.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.3.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.3.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.4.k_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.4.v_cache: torch.Size([1, 8, 347, 128]) -> torch.Size([1, 1, 347, 1024]) - layer.4.output: torch.Size([1, 347, 4096]) -> torch.Size([1, 1, 347, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,512B, BPFP=0.0566 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,648B, BPFP=0.0596 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,636B, BPFP=0.0819 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,076B, BPFP=0.0693 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,896B, BPFP=0.0877 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,296B, BPFP=0.0742 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,048B, BPFP=0.0911 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,488B, BPFP=0.0785 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,508B, BPFP=0.0790 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,396B, BPFP=0.0765 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,808B, BPFP=0.0552 -⌛️ [2/4] FRONTEND: Frontend time: 0.408s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 347, 128]) - layer.0.v_cache: torch.Size([1, 8, 347, 128]) - layer.1.k_cache: torch.Size([1, 8, 347, 128]) - layer.1.v_cache: torch.Size([1, 8, 347, 128]) - layer.2.k_cache: torch.Size([1, 8, 347, 128]) - layer.2.v_cache: torch.Size([1, 8, 347, 128]) - layer.3.k_cache: torch.Size([1, 8, 347, 128]) - layer.3.v_cache: torch.Size([1, 8, 347, 128]) - layer.4.k_cache: torch.Size([1, 8, 347, 128]) - layer.4.v_cache: torch.Size([1, 8, 347, 128]) - layer.4.output: torch.Size([1, 347, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.666s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 347, 128]) - layer.0.v_cache: torch.Size([1, 8, 347, 128]) - layer.1.k_cache: torch.Size([1, 8, 347, 128]) - layer.1.v_cache: torch.Size([1, 8, 347, 128]) - layer.2.k_cache: torch.Size([1, 8, 347, 128]) - layer.2.v_cache: torch.Size([1, 8, 347, 128]) - layer.3.k_cache: torch.Size([1, 8, 347, 128]) - layer.3.v_cache: torch.Size([1, 8, 347, 128]) - layer.4.k_cache: torch.Size([1, 8, 347, 128]) - layer.4.v_cache: torch.Size([1, 8, 347, 128]) - layer.4.output: torch.Size([1, 347, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02341881 58.43059370 - layer.0.v_cache 0.00000026 0.00066857 - layer.1.k_cache 0.00293388 5.67553570 - layer.1.v_cache 0.00000079 0.00284335 - layer.2.k_cache 0.00115427 2.08933366 - layer.2.v_cache 0.00000115 0.00428591 - layer.3.k_cache 0.00129772 2.42443584 - layer.3.v_cache 0.00000217 0.00723042 - layer.4.k_cache 0.00353122 6.27610264 - layer.4.v_cache 0.00000331 0.01223500 - layer.4.output 0.00014441 0.18908373 - ------------------------------------------------------------------------------------- - TOTAL 0.00235152 5.40568569 - (elements=4,974,592) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4974592 -Total Bytes 43312 -BPFP 0.0697 bits/point -EBPFP 0.1393 equivalent bits/point -MSE 5.405686 ----------------------- -------------------------------------------------------- -Time: 1.091s Load: 0.017s, Pack+Encode: 0.408s, Decode+Unpack: 0.666s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 347, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 347, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.4057 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample16-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample16-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample165-layer4-item1.zst (19/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample165-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 337, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 337, 128) -Output shape: (1, 337, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.output: torch.Size([1, 337, 4096]) -> torch.Size([1, 1, 337, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,560B, BPFP=0.0593 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,776B, BPFP=0.0644 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,816B, BPFP=0.0885 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,160B, BPFP=0.0733 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,020B, BPFP=0.0932 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,488B, BPFP=0.0809 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,208B, BPFP=0.0976 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,656B, BPFP=0.0848 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,636B, BPFP=0.0843 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,584B, BPFP=0.0831 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,672B, BPFP=0.0619 -⌛️ [2/4] FRONTEND: Frontend time: 0.403s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.output: torch.Size([1, 337, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.660s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.output: torch.Size([1, 337, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02403486 49.24318435 - layer.0.v_cache 0.00000026 0.00064193 - layer.1.k_cache 0.00293885 5.94665455 - layer.1.v_cache 0.00000076 0.00245237 - layer.2.k_cache 0.00113696 2.03075266 - layer.2.v_cache 0.00000113 0.00383637 - layer.3.k_cache 0.00132482 2.39616145 - layer.3.v_cache 0.00000208 0.00676907 - layer.4.k_cache 0.00376518 6.60762545 - layer.4.v_cache 0.00000304 0.01114338 - layer.4.output 0.00014776 0.20791861 - ------------------------------------------------------------------------------------- - TOTAL 0.00241421 4.79149257 - (elements=4,831,232) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4831232 -Total Bytes 45576 -BPFP 0.0755 bits/point -EBPFP 0.1509 equivalent bits/point -MSE 4.791493 ----------------------- -------------------------------------------------------- -Time: 1.080s Load: 0.017s, Pack+Encode: 0.403s, Decode+Unpack: 0.660s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 337, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7915 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample165-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample165-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample17-layer4-item1.zst (20/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample17-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 351, 128) -Output shape: (1, 351, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.output: torch.Size([1, 351, 4096]) -> torch.Size([1, 1, 351, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,508B, BPFP=0.0558 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,612B, BPFP=0.0581 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,644B, BPFP=0.0811 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,048B, BPFP=0.0678 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,876B, BPFP=0.0863 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,216B, BPFP=0.0716 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,060B, BPFP=0.0904 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,448B, BPFP=0.0767 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,524B, BPFP=0.0784 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,388B, BPFP=0.0754 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,704B, BPFP=0.0540 -⌛️ [2/4] FRONTEND: Frontend time: 0.413s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.661s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02526469 50.77270299 - layer.0.v_cache 0.00000026 0.00065910 - layer.1.k_cache 0.00294286 5.79491422 - layer.1.v_cache 0.00000076 0.00266311 - layer.2.k_cache 0.00116394 2.09165655 - layer.2.v_cache 0.00000112 0.00416304 - layer.3.k_cache 0.00133268 2.41553517 - layer.3.v_cache 0.00000207 0.00693897 - layer.4.k_cache 0.00360177 6.36015820 - layer.4.v_cache 0.00000315 0.01165357 - layer.4.output 0.00019540 0.19655292 - ------------------------------------------------------------------------------------- - TOTAL 0.00250678 4.87480404 - (elements=5,031,936) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5031936 -Total Bytes 43028 -BPFP 0.0684 bits/point -EBPFP 0.1368 equivalent bits/point -MSE 4.874804 ----------------------- -------------------------------------------------------- -Time: 1.092s Load: 0.018s, Pack+Encode: 0.413s, Decode+Unpack: 0.661s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8748 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample17-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample17-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample18-layer4-item1.zst (21/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample18-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 346, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 346, 128) -Output shape: (1, 346, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) -> torch.Size([1, 1, 346, 1024]) - layer.4.output: torch.Size([1, 346, 4096]) -> torch.Size([1, 1, 346, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,544B, BPFP=0.0574 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,696B, BPFP=0.0609 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,680B, BPFP=0.0831 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,144B, BPFP=0.0710 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,944B, BPFP=0.0891 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,384B, BPFP=0.0764 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,088B, BPFP=0.0923 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,584B, BPFP=0.0809 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,568B, BPFP=0.0806 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,476B, BPFP=0.0785 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,380B, BPFP=0.0586 -⌛️ [2/4] FRONTEND: Frontend time: 0.423s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.output: torch.Size([1, 346, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.697s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 346, 128]) - layer.0.v_cache: torch.Size([1, 8, 346, 128]) - layer.1.k_cache: torch.Size([1, 8, 346, 128]) - layer.1.v_cache: torch.Size([1, 8, 346, 128]) - layer.2.k_cache: torch.Size([1, 8, 346, 128]) - layer.2.v_cache: torch.Size([1, 8, 346, 128]) - layer.3.k_cache: torch.Size([1, 8, 346, 128]) - layer.3.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.k_cache: torch.Size([1, 8, 346, 128]) - layer.4.v_cache: torch.Size([1, 8, 346, 128]) - layer.4.output: torch.Size([1, 346, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02293222 61.47783824 - layer.0.v_cache 0.00000026 0.00064992 - layer.1.k_cache 0.00295457 5.78781992 - layer.1.v_cache 0.00000081 0.00283387 - layer.2.k_cache 0.00115565 2.06915865 - layer.2.v_cache 0.00000119 0.00443248 - layer.3.k_cache 0.00131728 2.40682295 - layer.3.v_cache 0.00000219 0.00743729 - layer.4.k_cache 0.00358690 6.38474375 - layer.4.v_cache 0.00000349 0.01219105 - layer.4.output 0.00015659 0.19546729 - ------------------------------------------------------------------------------------- - TOTAL 0.00232721 5.63827124 - (elements=4,960,256) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4960256 -Total Bytes 44488 -BPFP 0.0718 bits/point -EBPFP 0.1435 equivalent bits/point -MSE 5.638271 ----------------------- -------------------------------------------------------- -Time: 1.138s Load: 0.019s, Pack+Encode: 0.423s, Decode+Unpack: 0.697s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 346, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 346, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.6383 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample18-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample18-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample19-layer4-item1.zst (22/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample19-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 340, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 340, 128) -Output shape: (1, 340, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.output: torch.Size([1, 340, 4096]) -> torch.Size([1, 1, 340, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,552B, BPFP=0.0586 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,852B, BPFP=0.0655 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,768B, BPFP=0.0866 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,360B, BPFP=0.0772 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,012B, BPFP=0.0922 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,488B, BPFP=0.0801 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,220B, BPFP=0.0970 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,720B, BPFP=0.0855 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,632B, BPFP=0.0835 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,624B, BPFP=0.0833 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,276B, BPFP=0.0590 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.output: torch.Size([1, 340, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.670s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.output: torch.Size([1, 340, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02361420 53.61720473 - layer.0.v_cache 0.00000026 0.00067831 - layer.1.k_cache 0.00281368 5.98134730 - layer.1.v_cache 0.00000080 0.00288273 - layer.2.k_cache 0.00120688 2.11949104 - layer.2.v_cache 0.00000119 0.00452958 - layer.3.k_cache 0.00132000 2.45773423 - layer.3.v_cache 0.00000216 0.00750493 - layer.4.k_cache 0.00354241 6.47942684 - layer.4.v_cache 0.00000378 0.01264733 - layer.4.output 0.00015030 0.19117162 - ------------------------------------------------------------------------------------- - TOTAL 0.00236476 5.10343811 - (elements=4,874,240) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4874240 -Total Bytes 45504 -BPFP 0.0747 bits/point -EBPFP 0.1494 equivalent bits/point -MSE 5.103438 ----------------------- -------------------------------------------------------- -Time: 1.101s Load: 0.018s, Pack+Encode: 0.412s, Decode+Unpack: 0.670s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 340, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1034 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample19-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample19-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample2-layer4-item1.zst (23/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample2-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,580B, BPFP=0.0586 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,752B, BPFP=0.0625 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,744B, BPFP=0.0850 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,224B, BPFP=0.0732 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,992B, BPFP=0.0907 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,388B, BPFP=0.0769 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,152B, BPFP=0.0943 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,636B, BPFP=0.0826 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,612B, BPFP=0.0820 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,576B, BPFP=0.0812 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,184B, BPFP=0.0578 -⌛️ [2/4] FRONTEND: Frontend time: 0.403s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.658s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02321494 55.05328511 - layer.0.v_cache 0.00000027 0.00065062 - layer.1.k_cache 0.00289516 5.81276614 - layer.1.v_cache 0.00000077 0.00272946 - layer.2.k_cache 0.00116570 2.07315525 - layer.2.v_cache 0.00000112 0.00423713 - layer.3.k_cache 0.00132982 2.42054181 - layer.3.v_cache 0.00000218 0.00735483 - layer.4.k_cache 0.00351457 6.32109957 - layer.4.v_cache 0.00000367 0.01247812 - layer.4.output 0.00015182 0.20236452 - ------------------------------------------------------------------------------------- - TOTAL 0.00233825 5.17983972 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 44840 -BPFP 0.0727 bits/point -EBPFP 0.1455 equivalent bits/point -MSE 5.179840 ----------------------- -------------------------------------------------------- -Time: 1.078s Load: 0.018s, Pack+Encode: 0.403s, Decode+Unpack: 0.658s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1798 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample2-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample2-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample20-layer4-item1.zst (24/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample20-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 337, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 337, 128) -Output shape: (1, 337, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) -> torch.Size([1, 1, 337, 1024]) - layer.4.output: torch.Size([1, 337, 4096]) -> torch.Size([1, 1, 337, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,540B, BPFP=0.0589 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,804B, BPFP=0.0650 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,804B, BPFP=0.0882 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,328B, BPFP=0.0772 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,036B, BPFP=0.0936 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,484B, BPFP=0.0808 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,220B, BPFP=0.0978 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,684B, BPFP=0.0854 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,640B, BPFP=0.0844 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,608B, BPFP=0.0836 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,944B, BPFP=0.0576 -⌛️ [2/4] FRONTEND: Frontend time: 0.410s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.output: torch.Size([1, 337, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.657s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 337, 128]) - layer.0.v_cache: torch.Size([1, 8, 337, 128]) - layer.1.k_cache: torch.Size([1, 8, 337, 128]) - layer.1.v_cache: torch.Size([1, 8, 337, 128]) - layer.2.k_cache: torch.Size([1, 8, 337, 128]) - layer.2.v_cache: torch.Size([1, 8, 337, 128]) - layer.3.k_cache: torch.Size([1, 8, 337, 128]) - layer.3.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.k_cache: torch.Size([1, 8, 337, 128]) - layer.4.v_cache: torch.Size([1, 8, 337, 128]) - layer.4.output: torch.Size([1, 337, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435875 47.99796284 - layer.0.v_cache 0.00000026 0.00066612 - layer.1.k_cache 0.00286757 5.82150205 - layer.1.v_cache 0.00000078 0.00280107 - layer.2.k_cache 0.00117745 2.10429966 - layer.2.v_cache 0.00000112 0.00425337 - layer.3.k_cache 0.00130673 2.44460833 - layer.3.v_cache 0.00000209 0.00712371 - layer.4.k_cache 0.00356837 6.36958471 - layer.4.v_cache 0.00000322 0.01202556 - layer.4.output 0.00014638 0.18797326 - ------------------------------------------------------------------------------------- - TOTAL 0.00241942 4.67976575 - (elements=4,831,232) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4831232 -Total Bytes 45092 -BPFP 0.0747 bits/point -EBPFP 0.1493 equivalent bits/point -MSE 4.679766 ----------------------- -------------------------------------------------------- -Time: 1.083s Load: 0.017s, Pack+Encode: 0.410s, Decode+Unpack: 0.657s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 337, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 337, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.6798 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample20-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample20-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample21-layer4-item1.zst (25/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample21-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 395, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.021s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 395, 128) -Output shape: (1, 395, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.0.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.1.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.1.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.2.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.2.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.3.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.3.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.4.k_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.4.v_cache: torch.Size([1, 8, 395, 128]) -> torch.Size([1, 1, 395, 1024]) - layer.4.output: torch.Size([1, 395, 4096]) -> torch.Size([1, 1, 395, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,968B, BPFP=0.0587 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 3,096B, BPFP=0.0612 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 4,356B, BPFP=0.0862 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,644B, BPFP=0.0721 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,644B, BPFP=0.0919 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,884B, BPFP=0.0768 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,824B, BPFP=0.0954 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 4,140B, BPFP=0.0819 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 4,148B, BPFP=0.0820 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 4,052B, BPFP=0.0801 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 11,440B, BPFP=0.0566 -⌛️ [2/4] FRONTEND: Frontend time: 0.513s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 395, 128]) - layer.0.v_cache: torch.Size([1, 8, 395, 128]) - layer.1.k_cache: torch.Size([1, 8, 395, 128]) - layer.1.v_cache: torch.Size([1, 8, 395, 128]) - layer.2.k_cache: torch.Size([1, 8, 395, 128]) - layer.2.v_cache: torch.Size([1, 8, 395, 128]) - layer.3.k_cache: torch.Size([1, 8, 395, 128]) - layer.3.v_cache: torch.Size([1, 8, 395, 128]) - layer.4.k_cache: torch.Size([1, 8, 395, 128]) - layer.4.v_cache: torch.Size([1, 8, 395, 128]) - layer.4.output: torch.Size([1, 395, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.744s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 395, 128]) - layer.0.v_cache: torch.Size([1, 8, 395, 128]) - layer.1.k_cache: torch.Size([1, 8, 395, 128]) - layer.1.v_cache: torch.Size([1, 8, 395, 128]) - layer.2.k_cache: torch.Size([1, 8, 395, 128]) - layer.2.v_cache: torch.Size([1, 8, 395, 128]) - layer.3.k_cache: torch.Size([1, 8, 395, 128]) - layer.3.v_cache: torch.Size([1, 8, 395, 128]) - layer.4.k_cache: torch.Size([1, 8, 395, 128]) - layer.4.v_cache: torch.Size([1, 8, 395, 128]) - layer.4.output: torch.Size([1, 395, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02333523 47.09870451 - layer.0.v_cache 0.00000027 0.00067103 - layer.1.k_cache 0.00291865 5.77236328 - layer.1.v_cache 0.00000080 0.00266964 - layer.2.k_cache 0.00114101 2.04168794 - layer.2.v_cache 0.00000113 0.00401043 - layer.3.k_cache 0.00133612 2.41509840 - layer.3.v_cache 0.00000221 0.00691532 - layer.4.k_cache 0.00365959 6.33104171 - layer.4.v_cache 0.00000309 0.01146563 - layer.4.output 0.00017803 0.20454982 - ------------------------------------------------------------------------------------- - TOTAL 0.00236502 4.60734480 - (elements=5,662,720) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5662720 -Total Bytes 51196 -BPFP 0.0723 bits/point -EBPFP 0.1447 equivalent bits/point -MSE 4.607345 ----------------------- -------------------------------------------------------- -Time: 1.278s Load: 0.021s, Pack+Encode: 0.513s, Decode+Unpack: 0.744s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 395, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 395, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.6073 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample21-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample21-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample22-layer4-item1.zst (26/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample22-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 348, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 348, 128) -Output shape: (1, 348, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.output: torch.Size([1, 348, 4096]) -> torch.Size([1, 1, 348, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,500B, BPFP=0.0561 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,644B, BPFP=0.0594 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,636B, BPFP=0.0816 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,072B, BPFP=0.0690 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,900B, BPFP=0.0876 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,256B, BPFP=0.0731 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,036B, BPFP=0.0906 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,468B, BPFP=0.0779 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,504B, BPFP=0.0787 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,388B, BPFP=0.0761 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,752B, BPFP=0.0547 -⌛️ [2/4] FRONTEND: Frontend time: 0.416s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.output: torch.Size([1, 348, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.660s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.output: torch.Size([1, 348, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02579716 49.32714283 - layer.0.v_cache 0.00000026 0.00067311 - layer.1.k_cache 0.00289751 5.83981955 - layer.1.v_cache 0.00000077 0.00279234 - layer.2.k_cache 0.00114660 2.05947753 - layer.2.v_cache 0.00000117 0.00431202 - layer.3.k_cache 0.00133042 2.42600329 - layer.3.v_cache 0.00000210 0.00711378 - layer.4.k_cache 0.00358099 6.41393201 - layer.4.v_cache 0.00000335 0.01207467 - layer.4.output 0.00013932 0.18882969 - ------------------------------------------------------------------------------------- - TOTAL 0.00252269 4.77490428 - (elements=4,988,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4988928 -Total Bytes 43156 -BPFP 0.0692 bits/point -EBPFP 0.1384 equivalent bits/point -MSE 4.774904 ----------------------- -------------------------------------------------------- -Time: 1.094s Load: 0.018s, Pack+Encode: 0.416s, Decode+Unpack: 0.660s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 348, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7749 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample22-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample22-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample23-layer4-item1.zst (27/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample23-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 316, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 316, 128) -Output shape: (1, 316, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.output: torch.Size([1, 316, 4096]) -> torch.Size([1, 1, 316, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,168B, BPFP=0.0536 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,328B, BPFP=0.0576 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,996B, BPFP=0.0741 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,604B, BPFP=0.0644 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,220B, BPFP=0.0796 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,720B, BPFP=0.0672 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,328B, BPFP=0.0823 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,916B, BPFP=0.0721 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,900B, BPFP=0.0717 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,872B, BPFP=0.0710 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 8,240B, BPFP=0.0509 -⌛️ [2/4] FRONTEND: Frontend time: 0.355s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.output: torch.Size([1, 316, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.555s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.output: torch.Size([1, 316, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02512236 43.67690306 - layer.0.v_cache 0.00000027 0.00065420 - layer.1.k_cache 0.00293329 5.69518193 - layer.1.v_cache 0.00000080 0.00272137 - layer.2.k_cache 0.00117467 2.09971059 - layer.2.v_cache 0.00000111 0.00421640 - layer.3.k_cache 0.00131319 2.40221762 - layer.3.v_cache 0.00000216 0.00729712 - layer.4.k_cache 0.00359311 6.46123804 - layer.4.v_cache 0.00000344 0.01202188 - layer.4.output 0.00015753 0.19772650 - ------------------------------------------------------------------------------------- - TOTAL 0.00248390 4.36807630 - (elements=4,530,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4530176 -Total Bytes 36292 -BPFP 0.0641 bits/point -EBPFP 0.1282 equivalent bits/point -MSE 4.368076 ----------------------- -------------------------------------------------------- -Time: 0.926s Load: 0.017s, Pack+Encode: 0.355s, Decode+Unpack: 0.555s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 316, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.3681 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample23-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample23-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample24-layer4-item1.zst (28/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample24-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 348, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 348, 128) -Output shape: (1, 348, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) -> torch.Size([1, 1, 348, 1024]) - layer.4.output: torch.Size([1, 348, 4096]) -> torch.Size([1, 1, 348, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,500B, BPFP=0.0561 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,604B, BPFP=0.0585 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,628B, BPFP=0.0814 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,008B, BPFP=0.0675 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,884B, BPFP=0.0872 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,204B, BPFP=0.0719 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,024B, BPFP=0.0903 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,468B, BPFP=0.0779 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,480B, BPFP=0.0781 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,360B, BPFP=0.0754 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,908B, BPFP=0.0556 -⌛️ [2/4] FRONTEND: Frontend time: 0.409s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.output: torch.Size([1, 348, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.663s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 348, 128]) - layer.0.v_cache: torch.Size([1, 8, 348, 128]) - layer.1.k_cache: torch.Size([1, 8, 348, 128]) - layer.1.v_cache: torch.Size([1, 8, 348, 128]) - layer.2.k_cache: torch.Size([1, 8, 348, 128]) - layer.2.v_cache: torch.Size([1, 8, 348, 128]) - layer.3.k_cache: torch.Size([1, 8, 348, 128]) - layer.3.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.k_cache: torch.Size([1, 8, 348, 128]) - layer.4.v_cache: torch.Size([1, 8, 348, 128]) - layer.4.output: torch.Size([1, 348, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02368600 50.02865145 - layer.0.v_cache 0.00000027 0.00066559 - layer.1.k_cache 0.00288004 5.76693270 - layer.1.v_cache 0.00000079 0.00281661 - layer.2.k_cache 0.00117070 2.09817680 - layer.2.v_cache 0.00000116 0.00440353 - layer.3.k_cache 0.00131678 2.44406549 - layer.3.v_cache 0.00000214 0.00732024 - layer.4.k_cache 0.00357349 6.37227236 - layer.4.v_cache 0.00000347 0.01225528 - layer.4.output 0.00013697 0.19133787 - ------------------------------------------------------------------------------------- - TOTAL 0.00237019 4.82163654 - (elements=4,988,928) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4988928 -Total Bytes 43068 -BPFP 0.0691 bits/point -EBPFP 0.1381 equivalent bits/point -MSE 4.821637 ----------------------- -------------------------------------------------------- -Time: 1.091s Load: 0.019s, Pack+Encode: 0.409s, Decode+Unpack: 0.663s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 348, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 348, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8216 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample24-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample24-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample25-layer4-item1.zst (29/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample25-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 345, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 345, 128) -Output shape: (1, 345, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.output: torch.Size([1, 345, 4096]) -> torch.Size([1, 1, 345, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,560B, BPFP=0.0580 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,704B, BPFP=0.0612 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,700B, BPFP=0.0838 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,120B, BPFP=0.0707 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,952B, BPFP=0.0895 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,336B, BPFP=0.0755 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,120B, BPFP=0.0933 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,536B, BPFP=0.0801 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,600B, BPFP=0.0815 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,452B, BPFP=0.0782 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,028B, BPFP=0.0568 -⌛️ [2/4] FRONTEND: Frontend time: 0.410s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.output: torch.Size([1, 345, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.663s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.output: torch.Size([1, 345, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02357864 62.61454370 - layer.0.v_cache 0.00000026 0.00065143 - layer.1.k_cache 0.00288937 5.75005732 - layer.1.v_cache 0.00000077 0.00279767 - layer.2.k_cache 0.00116899 2.10311049 - layer.2.v_cache 0.00000115 0.00441825 - layer.3.k_cache 0.00131620 2.42499682 - layer.3.v_cache 0.00000216 0.00739048 - layer.4.k_cache 0.00353981 6.42227482 - layer.4.v_cache 0.00000338 0.01238594 - layer.4.output 0.00014567 0.19186914 - ------------------------------------------------------------------------------------- - TOTAL 0.00236310 5.72215025 - (elements=4,945,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4945920 -Total Bytes 44108 -BPFP 0.0713 bits/point -EBPFP 0.1427 equivalent bits/point -MSE 5.722150 ----------------------- -------------------------------------------------------- -Time: 1.091s Load: 0.019s, Pack+Encode: 0.410s, Decode+Unpack: 0.663s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 345, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.7222 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample25-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample25-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample26-layer4-item1.zst (30/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample26-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,620B, BPFP=0.0620 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,848B, BPFP=0.0674 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,800B, BPFP=0.0900 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,332B, BPFP=0.0789 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,096B, BPFP=0.0970 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,520B, BPFP=0.0833 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,192B, BPFP=0.0992 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,744B, BPFP=0.0886 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,624B, BPFP=0.0858 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,624B, BPFP=0.0858 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,384B, BPFP=0.0615 -⌛️ [2/4] FRONTEND: Frontend time: 0.407s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.660s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02434325 51.82829072 - layer.0.v_cache 0.00000027 0.00067571 - layer.1.k_cache 0.00287412 5.90469564 - layer.1.v_cache 0.00000081 0.00280709 - layer.2.k_cache 0.00119273 2.09731926 - layer.2.v_cache 0.00000117 0.00446316 - layer.3.k_cache 0.00130297 2.45067157 - layer.3.v_cache 0.00000221 0.00759112 - layer.4.k_cache 0.00352071 6.37182765 - layer.4.v_cache 0.00000330 0.01260852 - layer.4.output 0.00014370 0.19640175 - ------------------------------------------------------------------------------------- - TOTAL 0.00241545 4.96189696 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 45784 -BPFP 0.0774 bits/point -EBPFP 0.1548 equivalent bits/point -MSE 4.961897 ----------------------- -------------------------------------------------------- -Time: 1.086s Load: 0.019s, Pack+Encode: 0.407s, Decode+Unpack: 0.660s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.9619 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample26-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample26-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample27-layer4-item1.zst (31/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample27-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 334, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 334, 128) -Output shape: (1, 334, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.output: torch.Size([1, 334, 4096]) -> torch.Size([1, 1, 334, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,588B, BPFP=0.0605 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,772B, BPFP=0.0648 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,824B, BPFP=0.0894 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,304B, BPFP=0.0773 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,048B, BPFP=0.0947 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,480B, BPFP=0.0814 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,204B, BPFP=0.0983 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,680B, BPFP=0.0861 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,620B, BPFP=0.0847 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,516B, BPFP=0.0822 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,464B, BPFP=0.0612 -⌛️ [2/4] FRONTEND: Frontend time: 0.407s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.output: torch.Size([1, 334, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.658s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.output: torch.Size([1, 334, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02363260 49.63074827 - layer.0.v_cache 0.00000026 0.00065893 - layer.1.k_cache 0.00290221 5.81063386 - layer.1.v_cache 0.00000079 0.00276830 - layer.2.k_cache 0.00116170 2.09409958 - layer.2.v_cache 0.00000115 0.00435392 - layer.3.k_cache 0.00132769 2.43848387 - layer.3.v_cache 0.00000211 0.00724154 - layer.4.k_cache 0.00362300 6.63046612 - layer.4.v_cache 0.00000352 0.01214235 - layer.4.output 0.00014243 0.20537855 - ------------------------------------------------------------------------------------- - TOTAL 0.00237320 4.81807935 - (elements=4,788,224) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4788224 -Total Bytes 45500 -BPFP 0.0760 bits/point -EBPFP 0.1520 equivalent bits/point -MSE 4.818079 ----------------------- -------------------------------------------------------- -Time: 1.082s Load: 0.018s, Pack+Encode: 0.407s, Decode+Unpack: 0.658s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 334, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8181 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample27-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample27-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample28-layer4-item1.zst (32/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample28-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,596B, BPFP=0.0615 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,788B, BPFP=0.0660 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,792B, BPFP=0.0898 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,296B, BPFP=0.0780 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,068B, BPFP=0.0963 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,480B, BPFP=0.0824 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,188B, BPFP=0.0991 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,764B, BPFP=0.0891 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,604B, BPFP=0.0853 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,624B, BPFP=0.0858 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,456B, BPFP=0.0619 -⌛️ [2/4] FRONTEND: Frontend time: 0.401s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.661s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02410887 53.51128670 - layer.0.v_cache 0.00000026 0.00067212 - layer.1.k_cache 0.00288475 5.86014256 - layer.1.v_cache 0.00000079 0.00279044 - layer.2.k_cache 0.00119186 2.10797545 - layer.2.v_cache 0.00000115 0.00435358 - layer.3.k_cache 0.00132245 2.45619340 - layer.3.v_cache 0.00000226 0.00748538 - layer.4.k_cache 0.00352035 6.40752249 - layer.4.v_cache 0.00000339 0.01232404 - layer.4.output 0.00014978 0.20449406 - ------------------------------------------------------------------------------------- - TOTAL 0.00240252 5.08490874 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 45656 -BPFP 0.0772 bits/point -EBPFP 0.1544 equivalent bits/point -MSE 5.084909 ----------------------- -------------------------------------------------------- -Time: 1.079s Load: 0.017s, Pack+Encode: 0.401s, Decode+Unpack: 0.661s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.0849 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample28-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample28-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample29-layer4-item1.zst (33/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample29-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,660B, BPFP=0.0632 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,848B, BPFP=0.0676 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,792B, BPFP=0.0900 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,288B, BPFP=0.0781 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,084B, BPFP=0.0970 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,452B, BPFP=0.0820 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,228B, BPFP=0.1004 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,688B, BPFP=0.0876 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,636B, BPFP=0.0863 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,532B, BPFP=0.0839 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,640B, BPFP=0.0632 -⌛️ [2/4] FRONTEND: Frontend time: 0.402s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.651s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02426959 46.16233200 - layer.0.v_cache 0.00000027 0.00067784 - layer.1.k_cache 0.00291730 5.81555844 - layer.1.v_cache 0.00000079 0.00272934 - layer.2.k_cache 0.00116725 2.07261829 - layer.2.v_cache 0.00000113 0.00421059 - layer.3.k_cache 0.00130327 2.43248937 - layer.3.v_cache 0.00000218 0.00729624 - layer.4.k_cache 0.00359671 6.34132640 - layer.4.v_cache 0.00000327 0.01240384 - layer.4.output 0.00014627 0.19678168 - ------------------------------------------------------------------------------------- - TOTAL 0.00241763 4.54562636 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 45848 -BPFP 0.0778 bits/point -EBPFP 0.1555 equivalent bits/point -MSE 4.545626 ----------------------- -------------------------------------------------------- -Time: 1.071s Load: 0.018s, Pack+Encode: 0.402s, Decode+Unpack: 0.651s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.5456 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample29-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample29-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample3-layer4-item1.zst (34/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample3-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 365, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 365, 128) -Output shape: (1, 365, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.0.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.1.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.1.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.2.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.2.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.3.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.3.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.4.k_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.4.v_cache: torch.Size([1, 8, 365, 128]) -> torch.Size([1, 1, 365, 1024]) - layer.4.output: torch.Size([1, 365, 4096]) -> torch.Size([1, 1, 365, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,772B, BPFP=0.0593 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,944B, BPFP=0.0630 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,888B, BPFP=0.0832 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,372B, BPFP=0.0722 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,092B, BPFP=0.0876 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,516B, BPFP=0.0753 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,292B, BPFP=0.0919 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,756B, BPFP=0.0804 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,768B, BPFP=0.0807 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,696B, BPFP=0.0791 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,248B, BPFP=0.0548 -⌛️ [2/4] FRONTEND: Frontend time: 0.403s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 365, 128]) - layer.0.v_cache: torch.Size([1, 8, 365, 128]) - layer.1.k_cache: torch.Size([1, 8, 365, 128]) - layer.1.v_cache: torch.Size([1, 8, 365, 128]) - layer.2.k_cache: torch.Size([1, 8, 365, 128]) - layer.2.v_cache: torch.Size([1, 8, 365, 128]) - layer.3.k_cache: torch.Size([1, 8, 365, 128]) - layer.3.v_cache: torch.Size([1, 8, 365, 128]) - layer.4.k_cache: torch.Size([1, 8, 365, 128]) - layer.4.v_cache: torch.Size([1, 8, 365, 128]) - layer.4.output: torch.Size([1, 365, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.658s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 365, 128]) - layer.0.v_cache: torch.Size([1, 8, 365, 128]) - layer.1.k_cache: torch.Size([1, 8, 365, 128]) - layer.1.v_cache: torch.Size([1, 8, 365, 128]) - layer.2.k_cache: torch.Size([1, 8, 365, 128]) - layer.2.v_cache: torch.Size([1, 8, 365, 128]) - layer.3.k_cache: torch.Size([1, 8, 365, 128]) - layer.3.v_cache: torch.Size([1, 8, 365, 128]) - layer.4.k_cache: torch.Size([1, 8, 365, 128]) - layer.4.v_cache: torch.Size([1, 8, 365, 128]) - layer.4.output: torch.Size([1, 365, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02487199 55.01232877 - layer.0.v_cache 0.00000027 0.00067256 - layer.1.k_cache 0.00290279 5.91087864 - layer.1.v_cache 0.00000079 0.00283935 - layer.2.k_cache 0.00116885 2.12564363 - layer.2.v_cache 0.00000114 0.00446479 - layer.3.k_cache 0.00131452 2.46632632 - layer.3.v_cache 0.00000216 0.00752527 - layer.4.k_cache 0.00360833 6.49103435 - layer.4.v_cache 0.00000353 0.01238354 - layer.4.output 0.00014951 0.19199633 - ------------------------------------------------------------------------------------- - TOTAL 0.00246232 5.20014875 - (elements=5,232,640) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5232640 -Total Bytes 46344 -BPFP 0.0709 bits/point -EBPFP 0.1417 equivalent bits/point -MSE 5.200149 ----------------------- -------------------------------------------------------- -Time: 1.080s Load: 0.019s, Pack+Encode: 0.403s, Decode+Unpack: 0.658s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 365, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 365, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.2001 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample3-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample3-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample30-layer4-item1.zst (35/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample30-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 335, 128) -Output shape: (1, 335, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.output: torch.Size([1, 335, 4096]) -> torch.Size([1, 1, 335, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,576B, BPFP=0.0601 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,764B, BPFP=0.0645 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,824B, BPFP=0.0892 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,324B, BPFP=0.0775 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,036B, BPFP=0.0941 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,504B, BPFP=0.0817 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,220B, BPFP=0.0984 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,724B, BPFP=0.0868 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,604B, BPFP=0.0840 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,628B, BPFP=0.0846 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,216B, BPFP=0.0596 -⌛️ [2/4] FRONTEND: Frontend time: 0.400s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.653s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02462482 50.68888176 - layer.0.v_cache 0.00000027 0.00066924 - layer.1.k_cache 0.00287179 5.84891959 - layer.1.v_cache 0.00000079 0.00289372 - layer.2.k_cache 0.00118919 2.16469471 - layer.2.v_cache 0.00000114 0.00440728 - layer.3.k_cache 0.00130718 2.45027147 - layer.3.v_cache 0.00000215 0.00744078 - layer.4.k_cache 0.00355403 6.55230294 - layer.4.v_cache 0.00000335 0.01219779 - layer.4.output 0.00013864 0.19385724 - ------------------------------------------------------------------------------------- - TOTAL 0.00243638 4.89343630 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 45420 -BPFP 0.0757 bits/point -EBPFP 0.1513 equivalent bits/point -MSE 4.893436 ----------------------- -------------------------------------------------------- -Time: 1.070s Load: 0.017s, Pack+Encode: 0.400s, Decode+Unpack: 0.653s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8934 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample30-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample30-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample31-layer4-item1.zst (36/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample31-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 335, 128) -Output shape: (1, 335, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.output: torch.Size([1, 335, 4096]) -> torch.Size([1, 1, 335, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,572B, BPFP=0.0600 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,792B, BPFP=0.0651 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,828B, BPFP=0.0893 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,300B, BPFP=0.0770 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,040B, BPFP=0.0942 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,440B, BPFP=0.0802 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,184B, BPFP=0.0976 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,656B, BPFP=0.0853 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,620B, BPFP=0.0844 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,572B, BPFP=0.0833 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,284B, BPFP=0.0600 -⌛️ [2/4] FRONTEND: Frontend time: 0.404s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.663s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02394044 50.00057136 - layer.0.v_cache 0.00000026 0.00065364 - layer.1.k_cache 0.00289851 5.82296051 - layer.1.v_cache 0.00000079 0.00276586 - layer.2.k_cache 0.00115963 2.12316166 - layer.2.v_cache 0.00000116 0.00440694 - layer.3.k_cache 0.00131431 2.46323570 - layer.3.v_cache 0.00000219 0.00745482 - layer.4.k_cache 0.00369786 6.46776906 - layer.4.v_cache 0.00000335 0.01209672 - layer.4.output 0.00014933 0.20085802 - ------------------------------------------------------------------------------------- - TOTAL 0.00240113 4.83632203 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 45288 -BPFP 0.0754 bits/point -EBPFP 0.1509 equivalent bits/point -MSE 4.836322 ----------------------- -------------------------------------------------------- -Time: 1.085s Load: 0.018s, Pack+Encode: 0.404s, Decode+Unpack: 0.663s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8363 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample31-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample31-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample32-layer4-item1.zst (37/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample32-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 339, 128) -Output shape: (1, 339, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.output: torch.Size([1, 339, 4096]) -> torch.Size([1, 1, 339, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,540B, BPFP=0.0585 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,816B, BPFP=0.0649 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,764B, BPFP=0.0867 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,252B, BPFP=0.0749 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,984B, BPFP=0.0918 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,472B, BPFP=0.0800 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,172B, BPFP=0.0961 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,688B, BPFP=0.0850 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,612B, BPFP=0.0832 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,620B, BPFP=0.0834 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,232B, BPFP=0.0590 -⌛️ [2/4] FRONTEND: Frontend time: 0.401s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.659s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02382689 51.04576881 - layer.0.v_cache 0.00000027 0.00068145 - layer.1.k_cache 0.00292607 5.95698372 - layer.1.v_cache 0.00000081 0.00278678 - layer.2.k_cache 0.00125336 2.12358341 - layer.2.v_cache 0.00000116 0.00444862 - layer.3.k_cache 0.00131849 2.44826433 - layer.3.v_cache 0.00000220 0.00755756 - layer.4.k_cache 0.00356446 6.42754137 - layer.4.v_cache 0.00000343 0.01241017 - layer.4.output 0.00016600 0.19971146 - ------------------------------------------------------------------------------------- - TOTAL 0.00239722 4.91634801 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 45152 -BPFP 0.0743 bits/point -EBPFP 0.1487 equivalent bits/point -MSE 4.916348 ----------------------- -------------------------------------------------------- -Time: 1.078s Load: 0.018s, Pack+Encode: 0.401s, Decode+Unpack: 0.659s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.9163 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample32-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample32-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample33-layer4-item1.zst (38/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample33-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 362, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 362, 128) -Output shape: (1, 362, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) -> torch.Size([1, 1, 362, 1024]) - layer.4.output: torch.Size([1, 362, 4096]) -> torch.Size([1, 1, 362, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,756B, BPFP=0.0595 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,896B, BPFP=0.0625 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,908B, BPFP=0.0843 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,352B, BPFP=0.0723 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,128B, BPFP=0.0891 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,528B, BPFP=0.0761 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,316B, BPFP=0.0931 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,744B, BPFP=0.0808 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,764B, BPFP=0.0812 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,648B, BPFP=0.0787 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,960B, BPFP=0.0537 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.output: torch.Size([1, 362, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.657s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 362, 128]) - layer.0.v_cache: torch.Size([1, 8, 362, 128]) - layer.1.k_cache: torch.Size([1, 8, 362, 128]) - layer.1.v_cache: torch.Size([1, 8, 362, 128]) - layer.2.k_cache: torch.Size([1, 8, 362, 128]) - layer.2.v_cache: torch.Size([1, 8, 362, 128]) - layer.3.k_cache: torch.Size([1, 8, 362, 128]) - layer.3.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.k_cache: torch.Size([1, 8, 362, 128]) - layer.4.v_cache: torch.Size([1, 8, 362, 128]) - layer.4.output: torch.Size([1, 362, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02338883 54.95194881 - layer.0.v_cache 0.00000028 0.00068344 - layer.1.k_cache 0.00287711 5.81883956 - layer.1.v_cache 0.00000077 0.00282593 - layer.2.k_cache 0.00119138 2.13254475 - layer.2.v_cache 0.00000112 0.00438306 - layer.3.k_cache 0.00131510 2.43380535 - layer.3.v_cache 0.00000211 0.00733498 - layer.4.k_cache 0.00352384 6.39361943 - layer.4.v_cache 0.00000335 0.01230903 - layer.4.output 0.00013081 0.17993777 - ------------------------------------------------------------------------------------- - TOTAL 0.00234480 5.17700325 - (elements=5,189,632) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5189632 -Total Bytes 46000 -BPFP 0.0709 bits/point -EBPFP 0.1418 equivalent bits/point -MSE 5.177003 ----------------------- -------------------------------------------------------- -Time: 1.082s Load: 0.019s, Pack+Encode: 0.406s, Decode+Unpack: 0.657s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 362, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 362, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1770 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample33-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample33-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample34-layer4-item1.zst (39/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample34-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 318, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 318, 128) -Output shape: (1, 318, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) -> torch.Size([1, 1, 318, 1024]) - layer.4.output: torch.Size([1, 318, 4096]) -> torch.Size([1, 1, 318, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,016B, BPFP=0.0495 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,084B, BPFP=0.0512 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,864B, BPFP=0.0704 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,400B, BPFP=0.0590 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,112B, BPFP=0.0765 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,532B, BPFP=0.0622 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,224B, BPFP=0.0792 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,764B, BPFP=0.0679 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,748B, BPFP=0.0675 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,676B, BPFP=0.0657 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 7,708B, BPFP=0.0473 -⌛️ [2/4] FRONTEND: Frontend time: 0.355s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.output: torch.Size([1, 318, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.560s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 318, 128]) - layer.0.v_cache: torch.Size([1, 8, 318, 128]) - layer.1.k_cache: torch.Size([1, 8, 318, 128]) - layer.1.v_cache: torch.Size([1, 8, 318, 128]) - layer.2.k_cache: torch.Size([1, 8, 318, 128]) - layer.2.v_cache: torch.Size([1, 8, 318, 128]) - layer.3.k_cache: torch.Size([1, 8, 318, 128]) - layer.3.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.k_cache: torch.Size([1, 8, 318, 128]) - layer.4.v_cache: torch.Size([1, 8, 318, 128]) - layer.4.output: torch.Size([1, 318, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02477463 39.30336699 - layer.0.v_cache 0.00000027 0.00065881 - layer.1.k_cache 0.00289381 5.51731709 - layer.1.v_cache 0.00000083 0.00281477 - layer.2.k_cache 0.00119129 2.06353702 - layer.2.v_cache 0.00000117 0.00443531 - layer.3.k_cache 0.00131631 2.37822258 - layer.3.v_cache 0.00000257 0.00761255 - layer.4.k_cache 0.00356231 6.32546632 - layer.4.v_cache 0.00000346 0.01252334 - layer.4.output 0.00015646 0.20353068 - ------------------------------------------------------------------------------------- - TOTAL 0.00245518 4.03071982 - (elements=4,558,848) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4558848 -Total Bytes 34128 -BPFP 0.0599 bits/point -EBPFP 0.1198 equivalent bits/point -MSE 4.030720 ----------------------- -------------------------------------------------------- -Time: 0.933s Load: 0.017s, Pack+Encode: 0.355s, Decode+Unpack: 0.560s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 318, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 318, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.0307 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample34-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample34-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample35-layer4-item1.zst (40/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample35-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 342, 128) -Output shape: (1, 342, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.output: torch.Size([1, 342, 4096]) -> torch.Size([1, 1, 342, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,612B, BPFP=0.0597 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,828B, BPFP=0.0646 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,776B, BPFP=0.0863 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,284B, BPFP=0.0750 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,992B, BPFP=0.0912 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,388B, BPFP=0.0774 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,172B, BPFP=0.0953 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,656B, BPFP=0.0835 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,636B, BPFP=0.0831 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,616B, BPFP=0.0826 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,304B, BPFP=0.0588 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.678s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02487079 65.79886810 - layer.0.v_cache 0.00000026 0.00066660 - layer.1.k_cache 0.00288341 5.87911434 - layer.1.v_cache 0.00000082 0.00282087 - layer.2.k_cache 0.00117177 2.09938558 - layer.2.v_cache 0.00000117 0.00438104 - layer.3.k_cache 0.00130255 2.45047818 - layer.3.v_cache 0.00000215 0.00730154 - layer.4.k_cache 0.00351771 6.30228750 - layer.4.v_cache 0.00000337 0.01215768 - layer.4.output 0.00013439 0.18465124 - ------------------------------------------------------------------------------------- - TOTAL 0.00244940 5.94971903 - (elements=4,902,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4902912 -Total Bytes 45264 -BPFP 0.0739 bits/point -EBPFP 0.1477 equivalent bits/point -MSE 5.949719 ----------------------- -------------------------------------------------------- -Time: 1.108s Load: 0.018s, Pack+Encode: 0.412s, Decode+Unpack: 0.678s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.9497 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample35-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample35-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample36-layer4-item1.zst (41/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample36-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 343, 128) -Output shape: (1, 343, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.output: torch.Size([1, 343, 4096]) -> torch.Size([1, 1, 343, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,592B, BPFP=0.0590 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,764B, BPFP=0.0630 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,744B, BPFP=0.0853 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,268B, BPFP=0.0744 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,964B, BPFP=0.0903 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,428B, BPFP=0.0781 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,156B, BPFP=0.0947 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,676B, BPFP=0.0837 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,664B, BPFP=0.0835 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,564B, BPFP=0.0812 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,964B, BPFP=0.0567 -⌛️ [2/4] FRONTEND: Frontend time: 0.413s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.662s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02787456 58.94059197 - layer.0.v_cache 0.00000027 0.00067905 - layer.1.k_cache 0.00288980 5.89759831 - layer.1.v_cache 0.00000078 0.00286690 - layer.2.k_cache 0.00118800 2.12112631 - layer.2.v_cache 0.00000116 0.00443237 - layer.3.k_cache 0.00131625 2.45348747 - layer.3.v_cache 0.00000215 0.00747158 - layer.4.k_cache 0.00354145 6.39715772 - layer.4.v_cache 0.00000355 0.01241961 - layer.4.output 0.00016614 0.19448683 - ------------------------------------------------------------------------------------- - TOTAL 0.00267732 5.47255562 - (elements=4,917,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4917248 -Total Bytes 44784 -BPFP 0.0729 bits/point -EBPFP 0.1457 equivalent bits/point -MSE 5.472556 ----------------------- -------------------------------------------------------- -Time: 1.091s Load: 0.017s, Pack+Encode: 0.413s, Decode+Unpack: 0.662s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.4726 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample36-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample36-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample37-layer4-item1.zst (42/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample37-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.023s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 333, 128) -Output shape: (1, 333, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.output: torch.Size([1, 333, 4096]) -> torch.Size([1, 1, 333, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,584B, BPFP=0.0606 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,788B, BPFP=0.0654 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,848B, BPFP=0.0903 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,260B, BPFP=0.0765 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,032B, BPFP=0.0946 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,500B, BPFP=0.0821 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,192B, BPFP=0.0983 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,692B, BPFP=0.0866 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,616B, BPFP=0.0848 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,540B, BPFP=0.0831 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,996B, BPFP=0.0586 -⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.680s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02325115 52.81445312 - layer.0.v_cache 0.00000026 0.00067248 - layer.1.k_cache 0.00292822 5.92912994 - layer.1.v_cache 0.00000083 0.00282000 - layer.2.k_cache 0.00115284 2.11948447 - layer.2.v_cache 0.00000118 0.00438431 - layer.3.k_cache 0.00132958 2.43862888 - layer.3.v_cache 0.00000212 0.00734146 - layer.4.k_cache 0.00351398 6.36370648 - layer.4.v_cache 0.00000364 0.01232157 - layer.4.output 0.00014544 0.19641189 - ------------------------------------------------------------------------------------- - TOTAL 0.00234040 5.03418502 - (elements=4,773,888) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4773888 -Total Bytes 45048 -BPFP 0.0755 bits/point -EBPFP 0.1510 equivalent bits/point -MSE 5.034185 ----------------------- -------------------------------------------------------- -Time: 1.123s Load: 0.023s, Pack+Encode: 0.420s, Decode+Unpack: 0.680s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.0342 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample37-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample37-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample38-layer4-item1.zst (43/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample38-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 356, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 356, 128) -Output shape: (1, 356, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.output: torch.Size([1, 356, 4096]) -> torch.Size([1, 1, 356, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,564B, BPFP=0.0563 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,740B, BPFP=0.0601 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,716B, BPFP=0.0815 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,140B, BPFP=0.0689 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,932B, BPFP=0.0863 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,324B, BPFP=0.0729 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,128B, BPFP=0.0906 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,584B, BPFP=0.0787 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,588B, BPFP=0.0787 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,464B, BPFP=0.0760 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,040B, BPFP=0.0551 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.output: torch.Size([1, 356, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.681s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.output: torch.Size([1, 356, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02485108 51.60150215 - layer.0.v_cache 0.00000027 0.00065870 - layer.1.k_cache 0.00289550 5.94452659 - layer.1.v_cache 0.00000079 0.00276665 - layer.2.k_cache 0.00117641 2.13635803 - layer.2.v_cache 0.00000113 0.00432160 - layer.3.k_cache 0.00131994 2.44290950 - layer.3.v_cache 0.00000217 0.00741401 - layer.4.k_cache 0.00362269 6.52634936 - layer.4.v_cache 0.00000343 0.01236121 - layer.4.output 0.00016041 0.19685077 - ------------------------------------------------------------------------------------- - TOTAL 0.00246536 4.96189792 - (elements=5,103,616) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5103616 -Total Bytes 44220 -BPFP 0.0693 bits/point -EBPFP 0.1386 equivalent bits/point -MSE 4.961898 ----------------------- -------------------------------------------------------- -Time: 1.105s Load: 0.018s, Pack+Encode: 0.406s, Decode+Unpack: 0.681s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 356, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.9619 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample38-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample38-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample39-layer4-item1.zst (44/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample39-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 331, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 331, 128) -Output shape: (1, 331, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.0.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.1.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.1.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.2.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.2.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.3.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.3.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.4.k_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.4.v_cache: torch.Size([1, 8, 331, 128]) -> torch.Size([1, 1, 331, 1024]) - layer.4.output: torch.Size([1, 331, 4096]) -> torch.Size([1, 1, 331, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,620B, BPFP=0.0618 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,824B, BPFP=0.0667 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,788B, BPFP=0.0894 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,340B, BPFP=0.0788 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,044B, BPFP=0.0954 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,544B, BPFP=0.0836 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,184B, BPFP=0.0988 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,788B, BPFP=0.0894 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,640B, BPFP=0.0859 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,628B, BPFP=0.0856 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,444B, BPFP=0.0616 -⌛️ [2/4] FRONTEND: Frontend time: 0.423s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 331, 128]) - layer.0.v_cache: torch.Size([1, 8, 331, 128]) - layer.1.k_cache: torch.Size([1, 8, 331, 128]) - layer.1.v_cache: torch.Size([1, 8, 331, 128]) - layer.2.k_cache: torch.Size([1, 8, 331, 128]) - layer.2.v_cache: torch.Size([1, 8, 331, 128]) - layer.3.k_cache: torch.Size([1, 8, 331, 128]) - layer.3.v_cache: torch.Size([1, 8, 331, 128]) - layer.4.k_cache: torch.Size([1, 8, 331, 128]) - layer.4.v_cache: torch.Size([1, 8, 331, 128]) - layer.4.output: torch.Size([1, 331, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.669s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 331, 128]) - layer.0.v_cache: torch.Size([1, 8, 331, 128]) - layer.1.k_cache: torch.Size([1, 8, 331, 128]) - layer.1.v_cache: torch.Size([1, 8, 331, 128]) - layer.2.k_cache: torch.Size([1, 8, 331, 128]) - layer.2.v_cache: torch.Size([1, 8, 331, 128]) - layer.3.k_cache: torch.Size([1, 8, 331, 128]) - layer.3.v_cache: torch.Size([1, 8, 331, 128]) - layer.4.k_cache: torch.Size([1, 8, 331, 128]) - layer.4.v_cache: torch.Size([1, 8, 331, 128]) - layer.4.output: torch.Size([1, 331, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02326727 49.75463793 - layer.0.v_cache 0.00000026 0.00066294 - layer.1.k_cache 0.00291335 5.86293385 - layer.1.v_cache 0.00000080 0.00281239 - layer.2.k_cache 0.00116981 2.11647536 - layer.2.v_cache 0.00000115 0.00435518 - layer.3.k_cache 0.00132745 2.44094673 - layer.3.v_cache 0.00000216 0.00727241 - layer.4.k_cache 0.00359183 6.47634713 - layer.4.v_cache 0.00000354 0.01209841 - layer.4.output 0.00014916 0.20139654 - ------------------------------------------------------------------------------------- - TOTAL 0.00234816 4.82029489 - (elements=4,745,216) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4745216 -Total Bytes 45844 -BPFP 0.0773 bits/point -EBPFP 0.1546 equivalent bits/point -MSE 4.820295 ----------------------- -------------------------------------------------------- -Time: 1.108s Load: 0.016s, Pack+Encode: 0.423s, Decode+Unpack: 0.669s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 331, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 331, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8203 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample39-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample39-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample4-layer4-item1.zst (45/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample4-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 345, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 345, 128) -Output shape: (1, 345, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) -> torch.Size([1, 1, 345, 1024]) - layer.4.output: torch.Size([1, 345, 4096]) -> torch.Size([1, 1, 345, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,556B, BPFP=0.0579 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,752B, BPFP=0.0623 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,684B, BPFP=0.0834 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,132B, BPFP=0.0709 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,968B, BPFP=0.0899 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,380B, BPFP=0.0765 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,116B, BPFP=0.0932 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,628B, BPFP=0.0822 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,596B, BPFP=0.0814 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,568B, BPFP=0.0808 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,468B, BPFP=0.0593 -⌛️ [2/4] FRONTEND: Frontend time: 0.405s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.output: torch.Size([1, 345, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.658s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 345, 128]) - layer.0.v_cache: torch.Size([1, 8, 345, 128]) - layer.1.k_cache: torch.Size([1, 8, 345, 128]) - layer.1.v_cache: torch.Size([1, 8, 345, 128]) - layer.2.k_cache: torch.Size([1, 8, 345, 128]) - layer.2.v_cache: torch.Size([1, 8, 345, 128]) - layer.3.k_cache: torch.Size([1, 8, 345, 128]) - layer.3.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.k_cache: torch.Size([1, 8, 345, 128]) - layer.4.v_cache: torch.Size([1, 8, 345, 128]) - layer.4.output: torch.Size([1, 345, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02346668 59.51653646 - layer.0.v_cache 0.00000026 0.00065907 - layer.1.k_cache 0.00293293 5.63345717 - layer.1.v_cache 0.00000076 0.00260773 - layer.2.k_cache 0.00115776 2.03793538 - layer.2.v_cache 0.00000114 0.00421864 - layer.3.k_cache 0.00132774 2.42178619 - layer.3.v_cache 0.00000222 0.00731386 - layer.4.k_cache 0.00355356 6.29227808 - layer.4.v_cache 0.00000324 0.01203920 - layer.4.output 0.00016540 0.21022682 - ------------------------------------------------------------------------------------- - TOTAL 0.00236485 5.48355279 - (elements=4,945,920) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4945920 -Total Bytes 44848 -BPFP 0.0725 bits/point -EBPFP 0.1451 equivalent bits/point -MSE 5.483553 ----------------------- -------------------------------------------------------- -Time: 1.081s Load: 0.017s, Pack+Encode: 0.405s, Decode+Unpack: 0.658s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 345, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 345, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.4836 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample4-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample4-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample40-layer4-item1.zst (46/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample40-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,580B, BPFP=0.0586 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,792B, BPFP=0.0634 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,716B, BPFP=0.0844 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,208B, BPFP=0.0729 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,976B, BPFP=0.0903 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,412B, BPFP=0.0775 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,132B, BPFP=0.0938 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,616B, BPFP=0.0821 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,624B, BPFP=0.0823 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,536B, BPFP=0.0803 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,268B, BPFP=0.0583 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.669s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02380104 54.96984012 - layer.0.v_cache 0.00000026 0.00066107 - layer.1.k_cache 0.00287155 5.74462713 - layer.1.v_cache 0.00000078 0.00274674 - layer.2.k_cache 0.00117748 2.08726501 - layer.2.v_cache 0.00000114 0.00426819 - layer.3.k_cache 0.00132414 2.42645459 - layer.3.v_cache 0.00000215 0.00731407 - layer.4.k_cache 0.00352424 6.39194009 - layer.4.v_cache 0.00000349 0.01221273 - layer.4.output 0.00015726 0.19648197 - ------------------------------------------------------------------------------------- - TOTAL 0.00238109 5.17380412 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 44860 -BPFP 0.0728 bits/point -EBPFP 0.1455 equivalent bits/point -MSE 5.173804 ----------------------- -------------------------------------------------------- -Time: 1.099s Load: 0.018s, Pack+Encode: 0.412s, Decode+Unpack: 0.669s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1738 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample40-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample40-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample41-layer4-item1.zst (47/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample41-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 339, 128) -Output shape: (1, 339, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.output: torch.Size([1, 339, 4096]) -> torch.Size([1, 1, 339, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,548B, BPFP=0.0587 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,804B, BPFP=0.0646 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,764B, BPFP=0.0867 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,260B, BPFP=0.0751 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,992B, BPFP=0.0920 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,484B, BPFP=0.0803 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,168B, BPFP=0.0961 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,728B, BPFP=0.0859 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,644B, BPFP=0.0840 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,636B, BPFP=0.0838 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,156B, BPFP=0.0585 -⌛️ [2/4] FRONTEND: Frontend time: 0.408s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.672s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02519707 49.20498594 - layer.0.v_cache 0.00000026 0.00067571 - layer.1.k_cache 0.00285639 5.85766169 - layer.1.v_cache 0.00000078 0.00283276 - layer.2.k_cache 0.00115969 2.10184157 - layer.2.v_cache 0.00000117 0.00439822 - layer.3.k_cache 0.00131371 2.45690288 - layer.3.v_cache 0.00000219 0.00739261 - layer.4.k_cache 0.00358480 6.40633858 - layer.4.v_cache 0.00000371 0.01232374 - layer.4.output 0.00015316 0.19771706 - ------------------------------------------------------------------------------------- - TOTAL 0.00248089 4.77473014 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 45184 -BPFP 0.0744 bits/point -EBPFP 0.1488 equivalent bits/point -MSE 4.774730 ----------------------- -------------------------------------------------------- -Time: 1.097s Load: 0.017s, Pack+Encode: 0.408s, Decode+Unpack: 0.672s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7747 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample41-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample41-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample42-layer4-item1.zst (48/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample42-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 316, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 316, 128) -Output shape: (1, 316, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) -> torch.Size([1, 1, 316, 1024]) - layer.4.output: torch.Size([1, 316, 4096]) -> torch.Size([1, 1, 316, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,160B, BPFP=0.0534 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,232B, BPFP=0.0552 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,980B, BPFP=0.0737 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,536B, BPFP=0.0627 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,212B, BPFP=0.0794 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,664B, BPFP=0.0659 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,312B, BPFP=0.0819 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,848B, BPFP=0.0704 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,872B, BPFP=0.0710 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,840B, BPFP=0.0702 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 8,176B, BPFP=0.0505 -⌛️ [2/4] FRONTEND: Frontend time: 0.363s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.output: torch.Size([1, 316, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.571s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 316, 128]) - layer.0.v_cache: torch.Size([1, 8, 316, 128]) - layer.1.k_cache: torch.Size([1, 8, 316, 128]) - layer.1.v_cache: torch.Size([1, 8, 316, 128]) - layer.2.k_cache: torch.Size([1, 8, 316, 128]) - layer.2.v_cache: torch.Size([1, 8, 316, 128]) - layer.3.k_cache: torch.Size([1, 8, 316, 128]) - layer.3.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.k_cache: torch.Size([1, 8, 316, 128]) - layer.4.v_cache: torch.Size([1, 8, 316, 128]) - layer.4.output: torch.Size([1, 316, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02408213 43.69593058 - layer.0.v_cache 0.00000026 0.00065833 - layer.1.k_cache 0.00292725 5.71751037 - layer.1.v_cache 0.00000082 0.00275946 - layer.2.k_cache 0.00117417 2.06418175 - layer.2.v_cache 0.00000116 0.00437557 - layer.3.k_cache 0.00131215 2.40632803 - layer.3.v_cache 0.00000217 0.00746771 - layer.4.k_cache 0.00352248 6.32828029 - layer.4.v_cache 0.00000354 0.01248710 - layer.4.output 0.00015546 0.20623456 - ------------------------------------------------------------------------------------- - TOTAL 0.00240343 4.36177982 - (elements=4,530,176) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4530176 -Total Bytes 35832 -BPFP 0.0633 bits/point -EBPFP 0.1266 equivalent bits/point -MSE 4.361780 ----------------------- -------------------------------------------------------- -Time: 0.949s Load: 0.016s, Pack+Encode: 0.363s, Decode+Unpack: 0.571s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 316, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 316, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.3618 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample42-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample42-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample43-layer4-item1.zst (49/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample43-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.021s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 343, 128) -Output shape: (1, 343, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) -> torch.Size([1, 1, 343, 1024]) - layer.4.output: torch.Size([1, 343, 4096]) -> torch.Size([1, 1, 343, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,604B, BPFP=0.0593 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,832B, BPFP=0.0645 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,764B, BPFP=0.0857 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,280B, BPFP=0.0747 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,012B, BPFP=0.0914 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,460B, BPFP=0.0788 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,172B, BPFP=0.0950 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,684B, BPFP=0.0839 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,656B, BPFP=0.0833 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,628B, BPFP=0.0826 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,328B, BPFP=0.0588 -⌛️ [2/4] FRONTEND: Frontend time: 0.409s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.663s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 343, 128]) - layer.0.v_cache: torch.Size([1, 8, 343, 128]) - layer.1.k_cache: torch.Size([1, 8, 343, 128]) - layer.1.v_cache: torch.Size([1, 8, 343, 128]) - layer.2.k_cache: torch.Size([1, 8, 343, 128]) - layer.2.v_cache: torch.Size([1, 8, 343, 128]) - layer.3.k_cache: torch.Size([1, 8, 343, 128]) - layer.3.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.k_cache: torch.Size([1, 8, 343, 128]) - layer.4.v_cache: torch.Size([1, 8, 343, 128]) - layer.4.output: torch.Size([1, 343, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02365684 61.11467065 - layer.0.v_cache 0.00000027 0.00067367 - layer.1.k_cache 0.00286131 5.86824093 - layer.1.v_cache 0.00000077 0.00283565 - layer.2.k_cache 0.00115688 2.12808895 - layer.2.v_cache 0.00000135 0.00445558 - layer.3.k_cache 0.00129955 2.42846431 - layer.3.v_cache 0.00000214 0.00741952 - layer.4.k_cache 0.00368945 6.51486411 - layer.4.v_cache 0.00000337 0.01235629 - layer.4.output 0.00015649 0.19365005 - ------------------------------------------------------------------------------------- - TOTAL 0.00237842 5.63261927 - (elements=4,917,248) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4917248 -Total Bytes 45420 -BPFP 0.0739 bits/point -EBPFP 0.1478 equivalent bits/point -MSE 5.632619 ----------------------- -------------------------------------------------------- -Time: 1.093s Load: 0.021s, Pack+Encode: 0.409s, Decode+Unpack: 0.663s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 343, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 343, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.6326 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample43-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample43-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample44-layer4-item1.zst (50/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample44-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 326, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 326, 128) -Output shape: (1, 326, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.0.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.1.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.1.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.2.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.2.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.3.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.3.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.4.k_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.4.v_cache: torch.Size([1, 8, 326, 128]) -> torch.Size([1, 1, 326, 1024]) - layer.4.output: torch.Size([1, 326, 4096]) -> torch.Size([1, 1, 326, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,620B, BPFP=0.0628 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,808B, BPFP=0.0673 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,752B, BPFP=0.0899 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,236B, BPFP=0.0775 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,036B, BPFP=0.0967 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,404B, BPFP=0.0816 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,204B, BPFP=0.1007 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,620B, BPFP=0.0868 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,592B, BPFP=0.0861 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,524B, BPFP=0.0845 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,416B, BPFP=0.0624 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 326, 128]) - layer.0.v_cache: torch.Size([1, 8, 326, 128]) - layer.1.k_cache: torch.Size([1, 8, 326, 128]) - layer.1.v_cache: torch.Size([1, 8, 326, 128]) - layer.2.k_cache: torch.Size([1, 8, 326, 128]) - layer.2.v_cache: torch.Size([1, 8, 326, 128]) - layer.3.k_cache: torch.Size([1, 8, 326, 128]) - layer.3.v_cache: torch.Size([1, 8, 326, 128]) - layer.4.k_cache: torch.Size([1, 8, 326, 128]) - layer.4.v_cache: torch.Size([1, 8, 326, 128]) - layer.4.output: torch.Size([1, 326, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.664s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 326, 128]) - layer.0.v_cache: torch.Size([1, 8, 326, 128]) - layer.1.k_cache: torch.Size([1, 8, 326, 128]) - layer.1.v_cache: torch.Size([1, 8, 326, 128]) - layer.2.k_cache: torch.Size([1, 8, 326, 128]) - layer.2.v_cache: torch.Size([1, 8, 326, 128]) - layer.3.k_cache: torch.Size([1, 8, 326, 128]) - layer.3.v_cache: torch.Size([1, 8, 326, 128]) - layer.4.k_cache: torch.Size([1, 8, 326, 128]) - layer.4.v_cache: torch.Size([1, 8, 326, 128]) - layer.4.output: torch.Size([1, 326, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02438200 54.35538487 - layer.0.v_cache 0.00000027 0.00066784 - layer.1.k_cache 0.00290854 5.78727750 - layer.1.v_cache 0.00000077 0.00265072 - layer.2.k_cache 0.00114310 2.06246218 - layer.2.v_cache 0.00000112 0.00413591 - layer.3.k_cache 0.00131875 2.42385022 - layer.3.v_cache 0.00000210 0.00703744 - layer.4.k_cache 0.00360283 6.50749272 - layer.4.v_cache 0.00000318 0.01170144 - layer.4.output 0.00014710 0.19731994 - ------------------------------------------------------------------------------------- - TOTAL 0.00242508 5.13942433 - (elements=4,673,536) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4673536 -Total Bytes 45212 -BPFP 0.0774 bits/point -EBPFP 0.1548 equivalent bits/point -MSE 5.139424 ----------------------- -------------------------------------------------------- -Time: 1.092s Load: 0.017s, Pack+Encode: 0.411s, Decode+Unpack: 0.664s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 326, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 326, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1394 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample44-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample44-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample45-layer4-item1.zst (51/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample45-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 335, 128) -Output shape: (1, 335, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.output: torch.Size([1, 335, 4096]) -> torch.Size([1, 1, 335, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,564B, BPFP=0.0598 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,748B, BPFP=0.0641 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,828B, BPFP=0.0893 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,296B, BPFP=0.0769 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,016B, BPFP=0.0937 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,420B, BPFP=0.0798 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,220B, BPFP=0.0984 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,616B, BPFP=0.0843 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,604B, BPFP=0.0840 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,524B, BPFP=0.0822 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,308B, BPFP=0.0601 -⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.674s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02422888 50.34594216 - layer.0.v_cache 0.00000026 0.00066862 - layer.1.k_cache 0.00291837 5.79856977 - layer.1.v_cache 0.00000081 0.00285037 - layer.2.k_cache 0.00116761 2.11684006 - layer.2.v_cache 0.00000118 0.00439621 - layer.3.k_cache 0.00133299 2.45824886 - layer.3.v_cache 0.00000217 0.00740689 - layer.4.k_cache 0.00352932 6.48441144 - layer.4.v_cache 0.00000340 0.01227270 - layer.4.output 0.00015509 0.20117055 - ------------------------------------------------------------------------------------- - TOTAL 0.00241467 4.85973495 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 45144 -BPFP 0.0752 bits/point -EBPFP 0.1504 equivalent bits/point -MSE 4.859735 ----------------------- -------------------------------------------------------- -Time: 1.110s Load: 0.018s, Pack+Encode: 0.418s, Decode+Unpack: 0.674s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8597 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample45-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample45-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample46-layer4-item1.zst (52/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample46-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 324, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 324, 128) -Output shape: (1, 324, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.0.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.1.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.1.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.2.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.2.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.3.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.3.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.4.k_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.4.v_cache: torch.Size([1, 8, 324, 128]) -> torch.Size([1, 1, 324, 1024]) - layer.4.output: torch.Size([1, 324, 4096]) -> torch.Size([1, 1, 324, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,644B, BPFP=0.0638 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,772B, BPFP=0.0668 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,720B, BPFP=0.0897 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,104B, BPFP=0.0748 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,980B, BPFP=0.0960 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,272B, BPFP=0.0789 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,124B, BPFP=0.0994 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,576B, BPFP=0.0862 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,552B, BPFP=0.0856 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,448B, BPFP=0.0831 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,376B, BPFP=0.0625 -⌛️ [2/4] FRONTEND: Frontend time: 0.409s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 324, 128]) - layer.0.v_cache: torch.Size([1, 8, 324, 128]) - layer.1.k_cache: torch.Size([1, 8, 324, 128]) - layer.1.v_cache: torch.Size([1, 8, 324, 128]) - layer.2.k_cache: torch.Size([1, 8, 324, 128]) - layer.2.v_cache: torch.Size([1, 8, 324, 128]) - layer.3.k_cache: torch.Size([1, 8, 324, 128]) - layer.3.v_cache: torch.Size([1, 8, 324, 128]) - layer.4.k_cache: torch.Size([1, 8, 324, 128]) - layer.4.v_cache: torch.Size([1, 8, 324, 128]) - layer.4.output: torch.Size([1, 324, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.669s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 324, 128]) - layer.0.v_cache: torch.Size([1, 8, 324, 128]) - layer.1.k_cache: torch.Size([1, 8, 324, 128]) - layer.1.v_cache: torch.Size([1, 8, 324, 128]) - layer.2.k_cache: torch.Size([1, 8, 324, 128]) - layer.2.v_cache: torch.Size([1, 8, 324, 128]) - layer.3.k_cache: torch.Size([1, 8, 324, 128]) - layer.3.v_cache: torch.Size([1, 8, 324, 128]) - layer.4.k_cache: torch.Size([1, 8, 324, 128]) - layer.4.v_cache: torch.Size([1, 8, 324, 128]) - layer.4.output: torch.Size([1, 324, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02313130 49.13932292 - layer.0.v_cache 0.00000026 0.00066260 - layer.1.k_cache 0.00295515 5.88078365 - layer.1.v_cache 0.00000079 0.00274823 - layer.2.k_cache 0.00117180 2.13433084 - layer.2.v_cache 0.00000115 0.00436248 - layer.3.k_cache 0.00132196 2.44629528 - layer.3.v_cache 0.00000215 0.00745572 - layer.4.k_cache 0.00355989 6.49227792 - layer.4.v_cache 0.00000355 0.01227077 - layer.4.output 0.00015612 0.19922812 - ------------------------------------------------------------------------------------- - TOTAL 0.00234089 4.77981592 - (elements=4,644,864) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4644864 -Total Bytes 44568 -BPFP 0.0768 bits/point -EBPFP 0.1535 equivalent bits/point -MSE 4.779816 ----------------------- -------------------------------------------------------- -Time: 1.094s Load: 0.016s, Pack+Encode: 0.409s, Decode+Unpack: 0.669s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 324, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 324, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7798 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample46-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample46-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample47-layer4-item1.zst (53/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample47-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 332, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 332, 128) -Output shape: (1, 332, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.0.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.1.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.1.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.2.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.2.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.3.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.3.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.4.k_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.4.v_cache: torch.Size([1, 8, 332, 128]) -> torch.Size([1, 1, 332, 1024]) - layer.4.output: torch.Size([1, 332, 4096]) -> torch.Size([1, 1, 332, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,604B, BPFP=0.0613 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,892B, BPFP=0.0681 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,816B, BPFP=0.0898 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,400B, BPFP=0.0800 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,044B, BPFP=0.0952 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,564B, BPFP=0.0839 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,200B, BPFP=0.0988 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,760B, BPFP=0.0885 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,612B, BPFP=0.0850 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,672B, BPFP=0.0864 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,464B, BPFP=0.0616 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 332, 128]) - layer.0.v_cache: torch.Size([1, 8, 332, 128]) - layer.1.k_cache: torch.Size([1, 8, 332, 128]) - layer.1.v_cache: torch.Size([1, 8, 332, 128]) - layer.2.k_cache: torch.Size([1, 8, 332, 128]) - layer.2.v_cache: torch.Size([1, 8, 332, 128]) - layer.3.k_cache: torch.Size([1, 8, 332, 128]) - layer.3.v_cache: torch.Size([1, 8, 332, 128]) - layer.4.k_cache: torch.Size([1, 8, 332, 128]) - layer.4.v_cache: torch.Size([1, 8, 332, 128]) - layer.4.output: torch.Size([1, 332, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.672s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 332, 128]) - layer.0.v_cache: torch.Size([1, 8, 332, 128]) - layer.1.k_cache: torch.Size([1, 8, 332, 128]) - layer.1.v_cache: torch.Size([1, 8, 332, 128]) - layer.2.k_cache: torch.Size([1, 8, 332, 128]) - layer.2.v_cache: torch.Size([1, 8, 332, 128]) - layer.3.k_cache: torch.Size([1, 8, 332, 128]) - layer.3.v_cache: torch.Size([1, 8, 332, 128]) - layer.4.k_cache: torch.Size([1, 8, 332, 128]) - layer.4.v_cache: torch.Size([1, 8, 332, 128]) - layer.4.output: torch.Size([1, 332, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02360621 54.33845538 - layer.0.v_cache 0.00000026 0.00065946 - layer.1.k_cache 0.00288168 5.87831869 - layer.1.v_cache 0.00000079 0.00283755 - layer.2.k_cache 0.00118772 2.13928811 - layer.2.v_cache 0.00000117 0.00451714 - layer.3.k_cache 0.00132341 2.45352191 - layer.3.v_cache 0.00000218 0.00756304 - layer.4.k_cache 0.00354949 6.41732898 - layer.4.v_cache 0.00000378 0.01263310 - layer.4.output 0.00018345 0.20298602 - ------------------------------------------------------------------------------------- - TOTAL 0.00237789 5.14764767 - (elements=4,759,552) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4759552 -Total Bytes 46028 -BPFP 0.0774 bits/point -EBPFP 0.1547 equivalent bits/point -MSE 5.147648 ----------------------- -------------------------------------------------------- -Time: 1.095s Load: 0.017s, Pack+Encode: 0.406s, Decode+Unpack: 0.672s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 332, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 332, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1476 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample47-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample47-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample48-layer4-item1.zst (54/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample48-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 319, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 319, 128) -Output shape: (1, 319, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.output: torch.Size([1, 319, 4096]) -> torch.Size([1, 1, 319, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,976B, BPFP=0.0484 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,068B, BPFP=0.0506 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,836B, BPFP=0.0695 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,388B, BPFP=0.0585 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,088B, BPFP=0.0756 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,504B, BPFP=0.0613 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,208B, BPFP=0.0786 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,720B, BPFP=0.0666 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,716B, BPFP=0.0665 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,656B, BPFP=0.0650 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 7,424B, BPFP=0.0455 -⌛️ [2/4] FRONTEND: Frontend time: 0.352s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.output: torch.Size([1, 319, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.559s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.output: torch.Size([1, 319, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02438342 38.66491661 - layer.0.v_cache 0.00000027 0.00067065 - layer.1.k_cache 0.00287626 5.51194376 - layer.1.v_cache 0.00000082 0.00280645 - layer.2.k_cache 0.00122126 2.04302385 - layer.2.v_cache 0.00000114 0.00432657 - layer.3.k_cache 0.00133159 2.37196164 - layer.3.v_cache 0.00000213 0.00719192 - layer.4.k_cache 0.00352266 6.13853144 - layer.4.v_cache 0.00000325 0.01201949 - layer.4.output 0.00013715 0.18379990 - ------------------------------------------------------------------------------------- - TOTAL 0.00242081 3.96375657 - (elements=4,573,184) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4573184 -Total Bytes 33584 -BPFP 0.0587 bits/point -EBPFP 0.1175 equivalent bits/point -MSE 3.963757 ----------------------- -------------------------------------------------------- -Time: 0.927s Load: 0.016s, Pack+Encode: 0.352s, Decode+Unpack: 0.559s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 319, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 3.9638 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample48-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample48-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample49-layer4-item1.zst (55/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample49-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 335, 128) -Output shape: (1, 335, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) -> torch.Size([1, 1, 335, 1024]) - layer.4.output: torch.Size([1, 335, 4096]) -> torch.Size([1, 1, 335, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,564B, BPFP=0.0598 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,860B, BPFP=0.0667 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,828B, BPFP=0.0893 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,324B, BPFP=0.0775 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,040B, BPFP=0.0942 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,512B, BPFP=0.0819 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,240B, BPFP=0.0989 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,728B, BPFP=0.0869 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,652B, BPFP=0.0852 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,656B, BPFP=0.0853 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,036B, BPFP=0.0585 -⌛️ [2/4] FRONTEND: Frontend time: 0.399s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.659s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 335, 128]) - layer.0.v_cache: torch.Size([1, 8, 335, 128]) - layer.1.k_cache: torch.Size([1, 8, 335, 128]) - layer.1.v_cache: torch.Size([1, 8, 335, 128]) - layer.2.k_cache: torch.Size([1, 8, 335, 128]) - layer.2.v_cache: torch.Size([1, 8, 335, 128]) - layer.3.k_cache: torch.Size([1, 8, 335, 128]) - layer.3.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.k_cache: torch.Size([1, 8, 335, 128]) - layer.4.v_cache: torch.Size([1, 8, 335, 128]) - layer.4.output: torch.Size([1, 335, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02322828 50.62574627 - layer.0.v_cache 0.00000026 0.00066737 - layer.1.k_cache 0.00283722 5.81934942 - layer.1.v_cache 0.00000078 0.00280139 - layer.2.k_cache 0.00117890 2.12304724 - layer.2.v_cache 0.00000113 0.00429036 - layer.3.k_cache 0.00131429 2.46705668 - layer.3.v_cache 0.00000212 0.00732964 - layer.4.k_cache 0.00357066 6.40821041 - layer.4.v_cache 0.00000341 0.01225650 - layer.4.output 0.00013949 0.18628152 - ------------------------------------------------------------------------------------- - TOTAL 0.00233536 4.87256295 - (elements=4,802,560) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4802560 -Total Bytes 45440 -BPFP 0.0757 bits/point -EBPFP 0.1514 equivalent bits/point -MSE 4.872563 ----------------------- -------------------------------------------------------- -Time: 1.075s Load: 0.017s, Pack+Encode: 0.399s, Decode+Unpack: 0.659s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 335, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 335, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8726 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample49-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample49-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample5-layer4-item1.zst (56/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample5-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 354, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 354, 128) -Output shape: (1, 354, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.0.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.1.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.1.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.2.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.2.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.3.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.3.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.4.k_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.4.v_cache: torch.Size([1, 8, 354, 128]) -> torch.Size([1, 1, 354, 1024]) - layer.4.output: torch.Size([1, 354, 4096]) -> torch.Size([1, 1, 354, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,504B, BPFP=0.0553 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,768B, BPFP=0.0611 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,644B, BPFP=0.0804 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,188B, BPFP=0.0704 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,944B, BPFP=0.0870 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,344B, BPFP=0.0738 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,084B, BPFP=0.0901 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,596B, BPFP=0.0794 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,540B, BPFP=0.0781 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,484B, BPFP=0.0769 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,128B, BPFP=0.0559 -⌛️ [2/4] FRONTEND: Frontend time: 0.403s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 354, 128]) - layer.0.v_cache: torch.Size([1, 8, 354, 128]) - layer.1.k_cache: torch.Size([1, 8, 354, 128]) - layer.1.v_cache: torch.Size([1, 8, 354, 128]) - layer.2.k_cache: torch.Size([1, 8, 354, 128]) - layer.2.v_cache: torch.Size([1, 8, 354, 128]) - layer.3.k_cache: torch.Size([1, 8, 354, 128]) - layer.3.v_cache: torch.Size([1, 8, 354, 128]) - layer.4.k_cache: torch.Size([1, 8, 354, 128]) - layer.4.v_cache: torch.Size([1, 8, 354, 128]) - layer.4.output: torch.Size([1, 354, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.655s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 354, 128]) - layer.0.v_cache: torch.Size([1, 8, 354, 128]) - layer.1.k_cache: torch.Size([1, 8, 354, 128]) - layer.1.v_cache: torch.Size([1, 8, 354, 128]) - layer.2.k_cache: torch.Size([1, 8, 354, 128]) - layer.2.v_cache: torch.Size([1, 8, 354, 128]) - layer.3.k_cache: torch.Size([1, 8, 354, 128]) - layer.3.v_cache: torch.Size([1, 8, 354, 128]) - layer.4.k_cache: torch.Size([1, 8, 354, 128]) - layer.4.v_cache: torch.Size([1, 8, 354, 128]) - layer.4.output: torch.Size([1, 354, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02475441 54.09074307 - layer.0.v_cache 0.00000027 0.00068454 - layer.1.k_cache 0.00288960 5.79500877 - layer.1.v_cache 0.00000085 0.00283548 - layer.2.k_cache 0.00116959 2.09606313 - layer.2.v_cache 0.00000115 0.00436446 - layer.3.k_cache 0.00132243 2.44608027 - layer.3.v_cache 0.00000218 0.00746775 - layer.4.k_cache 0.00355194 6.45608848 - layer.4.v_cache 0.00000329 0.01242812 - layer.4.output 0.00015235 0.20469767 - ------------------------------------------------------------------------------------- - TOTAL 0.00245037 5.12361105 - (elements=5,074,944) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5074944 -Total Bytes 44224 -BPFP 0.0697 bits/point -EBPFP 0.1394 equivalent bits/point -MSE 5.123611 ----------------------- -------------------------------------------------------- -Time: 1.077s Load: 0.019s, Pack+Encode: 0.403s, Decode+Unpack: 0.655s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 354, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 354, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1236 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample5-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample5-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample50-layer4-item1.zst (57/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample50-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 327, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 327, 128) -Output shape: (1, 327, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.output: torch.Size([1, 327, 4096]) -> torch.Size([1, 1, 327, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,636B, BPFP=0.0630 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,788B, BPFP=0.0666 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,760B, BPFP=0.0898 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,292B, BPFP=0.0787 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,040B, BPFP=0.0965 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,496B, BPFP=0.0835 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,180B, BPFP=0.0999 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,716B, BPFP=0.0888 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,620B, BPFP=0.0865 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,616B, BPFP=0.0864 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,940B, BPFP=0.0653 -⌛️ [2/4] FRONTEND: Frontend time: 0.405s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.output: torch.Size([1, 327, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.654s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.output: torch.Size([1, 327, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02504845 53.83115921 - layer.0.v_cache 0.00000027 0.00067460 - layer.1.k_cache 0.00290162 5.75777181 - layer.1.v_cache 0.00000083 0.00281435 - layer.2.k_cache 0.00115956 2.09925978 - layer.2.v_cache 0.00000120 0.00440572 - layer.3.k_cache 0.00131802 2.44194418 - layer.3.v_cache 0.00000223 0.00746841 - layer.4.k_cache 0.00351161 6.34715304 - layer.4.v_cache 0.00000355 0.01239400 - layer.4.output 0.00016630 0.21176108 - ------------------------------------------------------------------------------------- - TOTAL 0.00247233 5.09657782 - (elements=4,687,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4687872 -Total Bytes 46084 -BPFP 0.0786 bits/point -EBPFP 0.1573 equivalent bits/point -MSE 5.096578 ----------------------- -------------------------------------------------------- -Time: 1.076s Load: 0.016s, Pack+Encode: 0.405s, Decode+Unpack: 0.654s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 327, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.0966 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample50-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample50-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample51-layer4-item1.zst (58/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample51-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 313, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 313, 128) -Output shape: (1, 313, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.0.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.1.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.1.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.2.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.2.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.3.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.3.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.4.k_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.4.v_cache: torch.Size([1, 8, 313, 128]) -> torch.Size([1, 1, 313, 1024]) - layer.4.output: torch.Size([1, 313, 4096]) -> torch.Size([1, 1, 313, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,276B, BPFP=0.0568 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,368B, BPFP=0.0591 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,216B, BPFP=0.0803 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,752B, BPFP=0.0687 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,412B, BPFP=0.0852 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,844B, BPFP=0.0710 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,516B, BPFP=0.0878 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,044B, BPFP=0.0760 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,116B, BPFP=0.0778 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,004B, BPFP=0.0750 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 8,184B, BPFP=0.0511 -⌛️ [2/4] FRONTEND: Frontend time: 0.347s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 313, 128]) - layer.0.v_cache: torch.Size([1, 8, 313, 128]) - layer.1.k_cache: torch.Size([1, 8, 313, 128]) - layer.1.v_cache: torch.Size([1, 8, 313, 128]) - layer.2.k_cache: torch.Size([1, 8, 313, 128]) - layer.2.v_cache: torch.Size([1, 8, 313, 128]) - layer.3.k_cache: torch.Size([1, 8, 313, 128]) - layer.3.v_cache: torch.Size([1, 8, 313, 128]) - layer.4.k_cache: torch.Size([1, 8, 313, 128]) - layer.4.v_cache: torch.Size([1, 8, 313, 128]) - layer.4.output: torch.Size([1, 313, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.561s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 313, 128]) - layer.0.v_cache: torch.Size([1, 8, 313, 128]) - layer.1.k_cache: torch.Size([1, 8, 313, 128]) - layer.1.v_cache: torch.Size([1, 8, 313, 128]) - layer.2.k_cache: torch.Size([1, 8, 313, 128]) - layer.2.v_cache: torch.Size([1, 8, 313, 128]) - layer.3.k_cache: torch.Size([1, 8, 313, 128]) - layer.3.v_cache: torch.Size([1, 8, 313, 128]) - layer.4.k_cache: torch.Size([1, 8, 313, 128]) - layer.4.v_cache: torch.Size([1, 8, 313, 128]) - layer.4.output: torch.Size([1, 313, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02366765 87.70167732 - layer.0.v_cache 0.00000027 0.00068714 - layer.1.k_cache 0.00295685 6.04727202 - layer.1.v_cache 0.00000081 0.00270786 - layer.2.k_cache 0.00117427 2.16458900 - layer.2.v_cache 0.00000111 0.00431267 - layer.3.k_cache 0.00131811 2.50453142 - layer.3.v_cache 0.00000214 0.00739660 - layer.4.k_cache 0.00360037 6.84338340 - layer.4.v_cache 0.00000322 0.01222145 - layer.4.output 0.00015721 0.21909284 - ------------------------------------------------------------------------------------- - TOTAL 0.00238240 7.58322502 - (elements=4,487,168) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4487168 -Total Bytes 37732 -BPFP 0.0673 bits/point -EBPFP 0.1345 equivalent bits/point -MSE 7.583225 ----------------------- -------------------------------------------------------- -Time: 0.924s Load: 0.016s, Pack+Encode: 0.347s, Decode+Unpack: 0.561s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 313, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 313, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 7.5832 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample51-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample51-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample52-layer4-item1.zst (59/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample52-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,656B, BPFP=0.0631 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,844B, BPFP=0.0675 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,804B, BPFP=0.0903 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,464B, BPFP=0.0823 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,092B, BPFP=0.0972 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,624B, BPFP=0.0861 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,208B, BPFP=0.0999 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,732B, BPFP=0.0886 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,644B, BPFP=0.0865 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,712B, BPFP=0.0881 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,780B, BPFP=0.0640 -⌛️ [2/4] FRONTEND: Frontend time: 0.401s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.662s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435147 45.63734506 - layer.0.v_cache 0.00000027 0.00067001 - layer.1.k_cache 0.00282437 5.87795121 - layer.1.v_cache 0.00000081 0.00282947 - layer.2.k_cache 0.00117800 2.10570102 - layer.2.v_cache 0.00000117 0.00436918 - layer.3.k_cache 0.00131145 2.45230668 - layer.3.v_cache 0.00000228 0.00748697 - layer.4.k_cache 0.00354228 6.40869215 - layer.4.v_cache 0.00000348 0.01248654 - layer.4.output 0.00016143 0.20350058 - ------------------------------------------------------------------------------------- - TOTAL 0.00241866 4.52313147 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 46560 -BPFP 0.0790 bits/point -EBPFP 0.1579 equivalent bits/point -MSE 4.523131 ----------------------- -------------------------------------------------------- -Time: 1.081s Load: 0.018s, Pack+Encode: 0.401s, Decode+Unpack: 0.662s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.5231 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample52-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample52-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample53-layer4-item1.zst (60/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample53-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,624B, BPFP=0.0621 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,792B, BPFP=0.0661 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,772B, BPFP=0.0893 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,352B, BPFP=0.0794 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,052B, BPFP=0.0959 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,492B, BPFP=0.0827 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,196B, BPFP=0.0993 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,688B, BPFP=0.0873 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,616B, BPFP=0.0856 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,540B, BPFP=0.0838 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,260B, BPFP=0.0607 -⌛️ [2/4] FRONTEND: Frontend time: 0.399s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.658s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02327405 50.92092803 - layer.0.v_cache 0.00000026 0.00067345 - layer.1.k_cache 0.00286502 6.00321526 - layer.1.v_cache 0.00000078 0.00283871 - layer.2.k_cache 0.00115867 2.14236543 - layer.2.v_cache 0.00000116 0.00440634 - layer.3.k_cache 0.00132713 2.47372178 - layer.3.v_cache 0.00000212 0.00739073 - layer.4.k_cache 0.00352212 6.41518851 - layer.4.v_cache 0.00000405 0.01256369 - layer.4.output 0.00014449 0.20043663 - ------------------------------------------------------------------------------------- - TOTAL 0.00233809 4.91321703 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 45384 -BPFP 0.0767 bits/point -EBPFP 0.1535 equivalent bits/point -MSE 4.913217 ----------------------- -------------------------------------------------------- -Time: 1.075s Load: 0.018s, Pack+Encode: 0.399s, Decode+Unpack: 0.658s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.9132 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample53-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample53-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample54-layer4-item1.zst (61/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample54-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 320, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 320, 128) -Output shape: (1, 320, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.output: torch.Size([1, 320, 4096]) -> torch.Size([1, 1, 320, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,916B, BPFP=0.0468 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,008B, BPFP=0.0490 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,824B, BPFP=0.0689 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,340B, BPFP=0.0571 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,076B, BPFP=0.0751 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,496B, BPFP=0.0609 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,180B, BPFP=0.0776 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,700B, BPFP=0.0659 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,688B, BPFP=0.0656 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,644B, BPFP=0.0646 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 7,300B, BPFP=0.0446 -⌛️ [2/4] FRONTEND: Frontend time: 0.347s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.output: torch.Size([1, 320, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.560s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.output: torch.Size([1, 320, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02461678 35.67826843 - layer.0.v_cache 0.00000026 0.00065282 - layer.1.k_cache 0.00291374 5.49006615 - layer.1.v_cache 0.00000082 0.00274141 - layer.2.k_cache 0.00121384 2.02615032 - layer.2.v_cache 0.00000117 0.00437261 - layer.3.k_cache 0.00132315 2.36714687 - layer.3.v_cache 0.00000219 0.00736274 - layer.4.k_cache 0.00353222 6.21736183 - layer.4.v_cache 0.00000326 0.01237880 - layer.4.output 0.00015660 0.19583064 - ------------------------------------------------------------------------------------- - TOTAL 0.00244527 3.75641604 - (elements=4,587,520) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4587520 -Total Bytes 33172 -BPFP 0.0578 bits/point -EBPFP 0.1157 equivalent bits/point -MSE 3.756416 ----------------------- -------------------------------------------------------- -Time: 0.924s Load: 0.017s, Pack+Encode: 0.347s, Decode+Unpack: 0.560s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 320, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 3.7564 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample54-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample54-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample55-layer4-item1.zst (62/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample55-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 339, 128) -Output shape: (1, 339, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.output: torch.Size([1, 339, 4096]) -> torch.Size([1, 1, 339, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,520B, BPFP=0.0581 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,756B, BPFP=0.0635 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,760B, BPFP=0.0867 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,264B, BPFP=0.0752 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,988B, BPFP=0.0919 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,436B, BPFP=0.0792 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,192B, BPFP=0.0966 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,692B, BPFP=0.0851 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,644B, BPFP=0.0840 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,608B, BPFP=0.0831 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,988B, BPFP=0.0575 -⌛️ [2/4] FRONTEND: Frontend time: 0.401s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.656s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02391626 50.18627858 - layer.0.v_cache 0.00000026 0.00066894 - layer.1.k_cache 0.00286983 5.82373299 - layer.1.v_cache 0.00000081 0.00283741 - layer.2.k_cache 0.00115809 2.09635831 - layer.2.v_cache 0.00000117 0.00441620 - layer.3.k_cache 0.00131372 2.43908583 - layer.3.v_cache 0.00000240 0.00749847 - layer.4.k_cache 0.00359509 6.36477562 - layer.4.v_cache 0.00000368 0.01249817 - layer.4.output 0.00014345 0.19201568 - ------------------------------------------------------------------------------------- - TOTAL 0.00238822 4.83615809 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 44848 -BPFP 0.0738 bits/point -EBPFP 0.1477 equivalent bits/point -MSE 4.836158 ----------------------- -------------------------------------------------------- -Time: 1.075s Load: 0.019s, Pack+Encode: 0.401s, Decode+Unpack: 0.656s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8362 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample55-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample55-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample56-layer4-item1.zst (63/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample56-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 314, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 314, 128) -Output shape: (1, 314, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.0.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.1.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.1.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.2.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.2.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.3.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.3.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.4.k_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.4.v_cache: torch.Size([1, 8, 314, 128]) -> torch.Size([1, 1, 314, 1024]) - layer.4.output: torch.Size([1, 314, 4096]) -> torch.Size([1, 1, 314, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,224B, BPFP=0.0553 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,308B, BPFP=0.0574 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,128B, BPFP=0.0778 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,712B, BPFP=0.0675 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,332B, BPFP=0.0829 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,808B, BPFP=0.0699 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,456B, BPFP=0.0860 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,984B, BPFP=0.0742 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,036B, BPFP=0.0755 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,976B, BPFP=0.0740 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 8,244B, BPFP=0.0513 -⌛️ [2/4] FRONTEND: Frontend time: 0.349s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 314, 128]) - layer.0.v_cache: torch.Size([1, 8, 314, 128]) - layer.1.k_cache: torch.Size([1, 8, 314, 128]) - layer.1.v_cache: torch.Size([1, 8, 314, 128]) - layer.2.k_cache: torch.Size([1, 8, 314, 128]) - layer.2.v_cache: torch.Size([1, 8, 314, 128]) - layer.3.k_cache: torch.Size([1, 8, 314, 128]) - layer.3.v_cache: torch.Size([1, 8, 314, 128]) - layer.4.k_cache: torch.Size([1, 8, 314, 128]) - layer.4.v_cache: torch.Size([1, 8, 314, 128]) - layer.4.output: torch.Size([1, 314, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.558s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 314, 128]) - layer.0.v_cache: torch.Size([1, 8, 314, 128]) - layer.1.k_cache: torch.Size([1, 8, 314, 128]) - layer.1.v_cache: torch.Size([1, 8, 314, 128]) - layer.2.k_cache: torch.Size([1, 8, 314, 128]) - layer.2.v_cache: torch.Size([1, 8, 314, 128]) - layer.3.k_cache: torch.Size([1, 8, 314, 128]) - layer.3.v_cache: torch.Size([1, 8, 314, 128]) - layer.4.k_cache: torch.Size([1, 8, 314, 128]) - layer.4.v_cache: torch.Size([1, 8, 314, 128]) - layer.4.output: torch.Size([1, 314, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02380536 62.96399781 - layer.0.v_cache 0.00000027 0.00068283 - layer.1.k_cache 0.00292133 5.93641070 - layer.1.v_cache 0.00000080 0.00281900 - layer.2.k_cache 0.00120652 2.13809301 - layer.2.v_cache 0.00000113 0.00435329 - layer.3.k_cache 0.00133526 2.46725036 - layer.3.v_cache 0.00000215 0.00740707 - layer.4.k_cache 0.00353655 6.60217207 - layer.4.v_cache 0.00000336 0.01232604 - layer.4.output 0.00016938 0.20514621 - ------------------------------------------------------------------------------------- - TOTAL 0.00239216 5.78257836 - (elements=4,501,504) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4501504 -Total Bytes 37208 -BPFP 0.0661 bits/point -EBPFP 0.1323 equivalent bits/point -MSE 5.782578 ----------------------- -------------------------------------------------------- -Time: 0.924s Load: 0.018s, Pack+Encode: 0.349s, Decode+Unpack: 0.558s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 314, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 314, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.7826 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample56-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample56-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample57-layer4-item1.zst (64/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample57-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.020s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,600B, BPFP=0.0590 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,836B, BPFP=0.0644 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,760B, BPFP=0.0854 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,308B, BPFP=0.0751 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,996B, BPFP=0.0908 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,496B, BPFP=0.0794 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,148B, BPFP=0.0942 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,660B, BPFP=0.0831 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,636B, BPFP=0.0826 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,552B, BPFP=0.0807 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,204B, BPFP=0.0579 -⌛️ [2/4] FRONTEND: Frontend time: 0.401s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.658s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02423059 56.86179937 - layer.0.v_cache 0.00000027 0.00068191 - layer.1.k_cache 0.00284490 5.86606225 - layer.1.v_cache 0.00000080 0.00286460 - layer.2.k_cache 0.00115895 2.09404276 - layer.2.v_cache 0.00000119 0.00449841 - layer.3.k_cache 0.00130300 2.44077549 - layer.3.v_cache 0.00000215 0.00752392 - layer.4.k_cache 0.00352154 6.31819188 - layer.4.v_cache 0.00000376 0.01253761 - layer.4.output 0.00013131 0.18162648 - ------------------------------------------------------------------------------------- - TOTAL 0.00239946 5.30967744 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 45196 -BPFP 0.0733 bits/point -EBPFP 0.1466 equivalent bits/point -MSE 5.309677 ----------------------- -------------------------------------------------------- -Time: 1.079s Load: 0.020s, Pack+Encode: 0.401s, Decode+Unpack: 0.658s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.3097 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample57-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample57-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample58-layer4-item1.zst (65/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample58-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 308, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 308, 128) -Output shape: (1, 308, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.0.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.1.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.1.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.2.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.2.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.3.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.3.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.4.k_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.4.v_cache: torch.Size([1, 8, 308, 128]) -> torch.Size([1, 1, 308, 1024]) - layer.4.output: torch.Size([1, 308, 4096]) -> torch.Size([1, 1, 308, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,380B, BPFP=0.0604 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,436B, BPFP=0.0618 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,348B, BPFP=0.0849 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,864B, BPFP=0.0726 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,492B, BPFP=0.0886 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,004B, BPFP=0.0762 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,652B, BPFP=0.0926 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,216B, BPFP=0.0816 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,232B, BPFP=0.0820 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,156B, BPFP=0.0801 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 8,920B, BPFP=0.0566 -⌛️ [2/4] FRONTEND: Frontend time: 0.355s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 308, 128]) - layer.0.v_cache: torch.Size([1, 8, 308, 128]) - layer.1.k_cache: torch.Size([1, 8, 308, 128]) - layer.1.v_cache: torch.Size([1, 8, 308, 128]) - layer.2.k_cache: torch.Size([1, 8, 308, 128]) - layer.2.v_cache: torch.Size([1, 8, 308, 128]) - layer.3.k_cache: torch.Size([1, 8, 308, 128]) - layer.3.v_cache: torch.Size([1, 8, 308, 128]) - layer.4.k_cache: torch.Size([1, 8, 308, 128]) - layer.4.v_cache: torch.Size([1, 8, 308, 128]) - layer.4.output: torch.Size([1, 308, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.562s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 308, 128]) - layer.0.v_cache: torch.Size([1, 8, 308, 128]) - layer.1.k_cache: torch.Size([1, 8, 308, 128]) - layer.1.v_cache: torch.Size([1, 8, 308, 128]) - layer.2.k_cache: torch.Size([1, 8, 308, 128]) - layer.2.v_cache: torch.Size([1, 8, 308, 128]) - layer.3.k_cache: torch.Size([1, 8, 308, 128]) - layer.3.v_cache: torch.Size([1, 8, 308, 128]) - layer.4.k_cache: torch.Size([1, 8, 308, 128]) - layer.4.v_cache: torch.Size([1, 8, 308, 128]) - layer.4.output: torch.Size([1, 308, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02345257 69.28539088 - layer.0.v_cache 0.00000027 0.00067709 - layer.1.k_cache 0.00291045 5.93277096 - layer.1.v_cache 0.00000082 0.00282555 - layer.2.k_cache 0.00116968 2.12145857 - layer.2.v_cache 0.00000115 0.00436655 - layer.3.k_cache 0.00131855 2.47197099 - layer.3.v_cache 0.00000224 0.00762131 - layer.4.k_cache 0.00354502 6.58912639 - layer.4.v_cache 0.00000324 0.01245181 - layer.4.output 0.00015550 0.20906912 - ------------------------------------------------------------------------------------- - TOTAL 0.00235900 6.23320976 - (elements=4,415,488) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4415488 -Total Bytes 39700 -BPFP 0.0719 bits/point -EBPFP 0.1439 equivalent bits/point -MSE 6.233210 ----------------------- -------------------------------------------------------- -Time: 0.932s Load: 0.016s, Pack+Encode: 0.355s, Decode+Unpack: 0.562s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 308, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 308, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.2332 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample58-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample58-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample59-layer4-item1.zst (66/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample59-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 339, 128) -Output shape: (1, 339, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) -> torch.Size([1, 1, 339, 1024]) - layer.4.output: torch.Size([1, 339, 4096]) -> torch.Size([1, 1, 339, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,552B, BPFP=0.0588 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,756B, BPFP=0.0635 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,768B, BPFP=0.0868 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,252B, BPFP=0.0749 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,984B, BPFP=0.0918 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,432B, BPFP=0.0791 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,164B, BPFP=0.0960 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,680B, BPFP=0.0848 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,636B, BPFP=0.0838 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,644B, BPFP=0.0840 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,380B, BPFP=0.0598 -⌛️ [2/4] FRONTEND: Frontend time: 0.401s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.652s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 339, 128]) - layer.0.v_cache: torch.Size([1, 8, 339, 128]) - layer.1.k_cache: torch.Size([1, 8, 339, 128]) - layer.1.v_cache: torch.Size([1, 8, 339, 128]) - layer.2.k_cache: torch.Size([1, 8, 339, 128]) - layer.2.v_cache: torch.Size([1, 8, 339, 128]) - layer.3.k_cache: torch.Size([1, 8, 339, 128]) - layer.3.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.k_cache: torch.Size([1, 8, 339, 128]) - layer.4.v_cache: torch.Size([1, 8, 339, 128]) - layer.4.output: torch.Size([1, 339, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02515205 49.48383343 - layer.0.v_cache 0.00000026 0.00066212 - layer.1.k_cache 0.00294413 5.85206518 - layer.1.v_cache 0.00000079 0.00274399 - layer.2.k_cache 0.00117757 2.09891926 - layer.2.v_cache 0.00000116 0.00432892 - layer.3.k_cache 0.00130963 2.42921074 - layer.3.v_cache 0.00000218 0.00744081 - layer.4.k_cache 0.00367572 6.53192985 - layer.4.v_cache 0.00000325 0.01220424 - layer.4.output 0.00014293 0.19519898 - ------------------------------------------------------------------------------------- - TOTAL 0.00248846 4.80029532 - (elements=4,859,904) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4859904 -Total Bytes 45248 -BPFP 0.0745 bits/point -EBPFP 0.1490 equivalent bits/point -MSE 4.800295 ----------------------- -------------------------------------------------------- -Time: 1.071s Load: 0.018s, Pack+Encode: 0.401s, Decode+Unpack: 0.652s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 339, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 339, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8003 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample59-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample59-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample6-layer4-item1.zst (67/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample6-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 355, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 355, 128) -Output shape: (1, 355, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.0.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.1.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.1.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.2.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.2.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.3.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.3.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.4.k_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.4.v_cache: torch.Size([1, 8, 355, 128]) -> torch.Size([1, 1, 355, 1024]) - layer.4.output: torch.Size([1, 355, 4096]) -> torch.Size([1, 1, 355, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,544B, BPFP=0.0560 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,780B, BPFP=0.0612 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,664B, BPFP=0.0806 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,212B, BPFP=0.0707 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,912B, BPFP=0.0861 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,348B, BPFP=0.0737 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,108B, BPFP=0.0904 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,612B, BPFP=0.0795 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,532B, BPFP=0.0777 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,440B, BPFP=0.0757 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,780B, BPFP=0.0538 -⌛️ [2/4] FRONTEND: Frontend time: 0.409s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 355, 128]) - layer.0.v_cache: torch.Size([1, 8, 355, 128]) - layer.1.k_cache: torch.Size([1, 8, 355, 128]) - layer.1.v_cache: torch.Size([1, 8, 355, 128]) - layer.2.k_cache: torch.Size([1, 8, 355, 128]) - layer.2.v_cache: torch.Size([1, 8, 355, 128]) - layer.3.k_cache: torch.Size([1, 8, 355, 128]) - layer.3.v_cache: torch.Size([1, 8, 355, 128]) - layer.4.k_cache: torch.Size([1, 8, 355, 128]) - layer.4.v_cache: torch.Size([1, 8, 355, 128]) - layer.4.output: torch.Size([1, 355, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.665s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 355, 128]) - layer.0.v_cache: torch.Size([1, 8, 355, 128]) - layer.1.k_cache: torch.Size([1, 8, 355, 128]) - layer.1.v_cache: torch.Size([1, 8, 355, 128]) - layer.2.k_cache: torch.Size([1, 8, 355, 128]) - layer.2.v_cache: torch.Size([1, 8, 355, 128]) - layer.3.k_cache: torch.Size([1, 8, 355, 128]) - layer.3.v_cache: torch.Size([1, 8, 355, 128]) - layer.4.k_cache: torch.Size([1, 8, 355, 128]) - layer.4.v_cache: torch.Size([1, 8, 355, 128]) - layer.4.output: torch.Size([1, 355, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02438693 57.72584727 - layer.0.v_cache 0.00000027 0.00067799 - layer.1.k_cache 0.00283684 6.02189357 - layer.1.v_cache 0.00000080 0.00287089 - layer.2.k_cache 0.00116829 2.10735585 - layer.2.v_cache 0.00000120 0.00448889 - layer.3.k_cache 0.00130356 2.45629866 - layer.3.v_cache 0.00000219 0.00749060 - layer.4.k_cache 0.00355311 6.53922136 - layer.4.v_cache 0.00000352 0.01243756 - layer.4.output 0.00013754 0.18397612 - ------------------------------------------------------------------------------------- - TOTAL 0.00241478 5.40103480 - (elements=5,089,280) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5089280 -Total Bytes 43932 -BPFP 0.0691 bits/point -EBPFP 0.1381 equivalent bits/point -MSE 5.401035 ----------------------- -------------------------------------------------------- -Time: 1.093s Load: 0.019s, Pack+Encode: 0.409s, Decode+Unpack: 0.665s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 355, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 355, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.4010 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample6-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample6-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample60-layer4-item1.zst (68/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample60-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 334, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 334, 128) -Output shape: (1, 334, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) -> torch.Size([1, 1, 334, 1024]) - layer.4.output: torch.Size([1, 334, 4096]) -> torch.Size([1, 1, 334, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,592B, BPFP=0.0606 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,812B, BPFP=0.0658 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,832B, BPFP=0.0896 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,340B, BPFP=0.0781 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,076B, BPFP=0.0953 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,520B, BPFP=0.0823 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,220B, BPFP=0.0987 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,732B, BPFP=0.0873 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,636B, BPFP=0.0850 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,648B, BPFP=0.0853 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,056B, BPFP=0.0588 -⌛️ [2/4] FRONTEND: Frontend time: 0.403s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.output: torch.Size([1, 334, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.662s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 334, 128]) - layer.0.v_cache: torch.Size([1, 8, 334, 128]) - layer.1.k_cache: torch.Size([1, 8, 334, 128]) - layer.1.v_cache: torch.Size([1, 8, 334, 128]) - layer.2.k_cache: torch.Size([1, 8, 334, 128]) - layer.2.v_cache: torch.Size([1, 8, 334, 128]) - layer.3.k_cache: torch.Size([1, 8, 334, 128]) - layer.3.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.k_cache: torch.Size([1, 8, 334, 128]) - layer.4.v_cache: torch.Size([1, 8, 334, 128]) - layer.4.output: torch.Size([1, 334, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02444427 51.63054945 - layer.0.v_cache 0.00000026 0.00067766 - layer.1.k_cache 0.00284576 5.96231024 - layer.1.v_cache 0.00000078 0.00285250 - layer.2.k_cache 0.00117801 2.13832558 - layer.2.v_cache 0.00000115 0.00436112 - layer.3.k_cache 0.00132187 2.45875668 - layer.3.v_cache 0.00000218 0.00753146 - layer.4.k_cache 0.00348374 6.34809629 - layer.4.v_cache 0.00000337 0.01256771 - layer.4.output 0.00014793 0.19269856 - ------------------------------------------------------------------------------------- - TOTAL 0.00241951 4.95263021 - (elements=4,788,224) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4788224 -Total Bytes 45464 -BPFP 0.0760 bits/point -EBPFP 0.1519 equivalent bits/point -MSE 4.952630 ----------------------- -------------------------------------------------------- -Time: 1.083s Load: 0.018s, Pack+Encode: 0.403s, Decode+Unpack: 0.662s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 334, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 334, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.9526 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample60-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample60-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample61-layer4-item1.zst (69/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample61-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 342, 128) -Output shape: (1, 342, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.output: torch.Size([1, 342, 4096]) -> torch.Size([1, 1, 342, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,572B, BPFP=0.0588 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,796B, BPFP=0.0639 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,768B, BPFP=0.0861 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,324B, BPFP=0.0759 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,972B, BPFP=0.0907 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,496B, BPFP=0.0799 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,172B, BPFP=0.0953 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,660B, BPFP=0.0836 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,644B, BPFP=0.0832 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,572B, BPFP=0.0816 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,840B, BPFP=0.0562 -⌛️ [2/4] FRONTEND: Frontend time: 0.400s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.666s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02429651 67.70795641 - layer.0.v_cache 0.00000027 0.00067336 - layer.1.k_cache 0.00293320 5.86661962 - layer.1.v_cache 0.00000079 0.00287803 - layer.2.k_cache 0.00117175 2.12927532 - layer.2.v_cache 0.00000114 0.00440357 - layer.3.k_cache 0.00133100 2.45247878 - layer.3.v_cache 0.00000223 0.00730842 - layer.4.k_cache 0.00354789 6.47051609 - layer.4.v_cache 0.00000344 0.01214605 - layer.4.output 0.00015145 0.19210293 - ------------------------------------------------------------------------------------- - TOTAL 0.00242100 6.10161910 - (elements=4,902,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4902912 -Total Bytes 44816 -BPFP 0.0731 bits/point -EBPFP 0.1463 equivalent bits/point -MSE 6.101619 ----------------------- -------------------------------------------------------- -Time: 1.083s Load: 0.017s, Pack+Encode: 0.400s, Decode+Unpack: 0.666s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.1016 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample61-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample61-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample62-layer4-item1.zst (70/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample62-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 312, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 312, 128) -Output shape: (1, 312, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.0.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.1.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.1.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.2.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.2.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.3.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.3.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.4.k_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.4.v_cache: torch.Size([1, 8, 312, 128]) -> torch.Size([1, 1, 312, 1024]) - layer.4.output: torch.Size([1, 312, 4096]) -> torch.Size([1, 1, 312, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,308B, BPFP=0.0578 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,392B, BPFP=0.0599 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,268B, BPFP=0.0818 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,764B, BPFP=0.0692 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,432B, BPFP=0.0859 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,928B, BPFP=0.0733 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,604B, BPFP=0.0902 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,160B, BPFP=0.0791 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,152B, BPFP=0.0789 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,092B, BPFP=0.0774 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 8,508B, BPFP=0.0533 -⌛️ [2/4] FRONTEND: Frontend time: 0.348s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 312, 128]) - layer.0.v_cache: torch.Size([1, 8, 312, 128]) - layer.1.k_cache: torch.Size([1, 8, 312, 128]) - layer.1.v_cache: torch.Size([1, 8, 312, 128]) - layer.2.k_cache: torch.Size([1, 8, 312, 128]) - layer.2.v_cache: torch.Size([1, 8, 312, 128]) - layer.3.k_cache: torch.Size([1, 8, 312, 128]) - layer.3.v_cache: torch.Size([1, 8, 312, 128]) - layer.4.k_cache: torch.Size([1, 8, 312, 128]) - layer.4.v_cache: torch.Size([1, 8, 312, 128]) - layer.4.output: torch.Size([1, 312, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.559s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 312, 128]) - layer.0.v_cache: torch.Size([1, 8, 312, 128]) - layer.1.k_cache: torch.Size([1, 8, 312, 128]) - layer.1.v_cache: torch.Size([1, 8, 312, 128]) - layer.2.k_cache: torch.Size([1, 8, 312, 128]) - layer.2.v_cache: torch.Size([1, 8, 312, 128]) - layer.3.k_cache: torch.Size([1, 8, 312, 128]) - layer.3.v_cache: torch.Size([1, 8, 312, 128]) - layer.4.k_cache: torch.Size([1, 8, 312, 128]) - layer.4.v_cache: torch.Size([1, 8, 312, 128]) - layer.4.output: torch.Size([1, 312, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02396220 108.74848508 - layer.0.v_cache 0.00000026 0.00067443 - layer.1.k_cache 0.00299662 6.27348719 - layer.1.v_cache 0.00000081 0.00277264 - layer.2.k_cache 0.00115233 2.16011654 - layer.2.v_cache 0.00000114 0.00438147 - layer.3.k_cache 0.00133293 2.51111661 - layer.3.v_cache 0.00000220 0.00747396 - layer.4.k_cache 0.00362032 6.96090933 - layer.4.v_cache 0.00000325 0.01208508 - layer.4.output 0.00015781 0.21980009 - ------------------------------------------------------------------------------------- - TOTAL 0.00240738 9.11147876 - (elements=4,472,832) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4472832 -Total Bytes 38608 -BPFP 0.0691 bits/point -EBPFP 0.1381 equivalent bits/point -MSE 9.111479 ----------------------- -------------------------------------------------------- -Time: 0.923s Load: 0.016s, Pack+Encode: 0.348s, Decode+Unpack: 0.559s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 312, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 312, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 9.1115 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample62-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample62-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample63-layer4-item1.zst (71/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample63-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 351, 128) -Output shape: (1, 351, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.output: torch.Size([1, 351, 4096]) -> torch.Size([1, 1, 351, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,500B, BPFP=0.0556 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,712B, BPFP=0.0604 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,660B, BPFP=0.0815 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,144B, BPFP=0.0700 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,888B, BPFP=0.0865 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,316B, BPFP=0.0738 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,060B, BPFP=0.0904 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,580B, BPFP=0.0797 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,516B, BPFP=0.0783 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,432B, BPFP=0.0764 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,788B, BPFP=0.0545 -⌛️ [2/4] FRONTEND: Frontend time: 0.401s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.660s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435836 49.44105012 - layer.0.v_cache 0.00000027 0.00067792 - layer.1.k_cache 0.00294291 5.69916763 - layer.1.v_cache 0.00000079 0.00286277 - layer.2.k_cache 0.00116231 2.07903253 - layer.2.v_cache 0.00000118 0.00439810 - layer.3.k_cache 0.00133045 2.43018222 - layer.3.v_cache 0.00000222 0.00745952 - layer.4.k_cache 0.00357138 6.40468847 - layer.4.v_cache 0.00000341 0.01223342 - layer.4.output 0.00014419 0.19324791 - ------------------------------------------------------------------------------------- - TOTAL 0.00242500 4.77533888 - (elements=5,031,936) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5031936 -Total Bytes 43596 -BPFP 0.0693 bits/point -EBPFP 0.1386 equivalent bits/point -MSE 4.775339 ----------------------- -------------------------------------------------------- -Time: 1.080s Load: 0.019s, Pack+Encode: 0.401s, Decode+Unpack: 0.660s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7753 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample63-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample63-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample64-layer4-item1.zst (72/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample64-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 323, 128) -Output shape: (1, 323, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.output: torch.Size([1, 323, 4096]) -> torch.Size([1, 1, 323, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,584B, BPFP=0.0625 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,648B, BPFP=0.0640 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,692B, BPFP=0.0893 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,132B, BPFP=0.0758 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,980B, BPFP=0.0963 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,252B, BPFP=0.0787 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,120B, BPFP=0.0997 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,520B, BPFP=0.0851 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,516B, BPFP=0.0850 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,364B, BPFP=0.0814 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,944B, BPFP=0.0601 -⌛️ [2/4] FRONTEND: Frontend time: 0.400s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.658s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02343196 65.41981908 - layer.0.v_cache 0.00000026 0.00066257 - layer.1.k_cache 0.00287105 5.89350102 - layer.1.v_cache 0.00000078 0.00282350 - layer.2.k_cache 0.00117289 2.12249246 - layer.2.v_cache 0.00000112 0.00429195 - layer.3.k_cache 0.00130124 2.45086538 - layer.3.v_cache 0.00000213 0.00717365 - layer.4.k_cache 0.00357998 6.56177892 - layer.4.v_cache 0.00000342 0.01210352 - layer.4.output 0.00014651 0.19509923 - ------------------------------------------------------------------------------------- - TOTAL 0.00235363 5.94685064 - (elements=4,630,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4630528 -Total Bytes 43752 -BPFP 0.0756 bits/point -EBPFP 0.1512 equivalent bits/point -MSE 5.946851 ----------------------- -------------------------------------------------------- -Time: 1.075s Load: 0.018s, Pack+Encode: 0.400s, Decode+Unpack: 0.658s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.9469 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample64-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample64-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample65-layer4-item1.zst (73/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample65-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 356, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 356, 128) -Output shape: (1, 356, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) -> torch.Size([1, 1, 356, 1024]) - layer.4.output: torch.Size([1, 356, 4096]) -> torch.Size([1, 1, 356, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,568B, BPFP=0.0564 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,788B, BPFP=0.0612 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,728B, BPFP=0.0818 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,284B, BPFP=0.0721 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,924B, BPFP=0.0861 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,392B, BPFP=0.0744 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,132B, BPFP=0.0907 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,596B, BPFP=0.0789 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,588B, BPFP=0.0787 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,508B, BPFP=0.0770 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,008B, BPFP=0.0549 -⌛️ [2/4] FRONTEND: Frontend time: 0.401s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.output: torch.Size([1, 356, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.670s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 356, 128]) - layer.0.v_cache: torch.Size([1, 8, 356, 128]) - layer.1.k_cache: torch.Size([1, 8, 356, 128]) - layer.1.v_cache: torch.Size([1, 8, 356, 128]) - layer.2.k_cache: torch.Size([1, 8, 356, 128]) - layer.2.v_cache: torch.Size([1, 8, 356, 128]) - layer.3.k_cache: torch.Size([1, 8, 356, 128]) - layer.3.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.k_cache: torch.Size([1, 8, 356, 128]) - layer.4.v_cache: torch.Size([1, 8, 356, 128]) - layer.4.output: torch.Size([1, 356, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02477347 50.65989839 - layer.0.v_cache 0.00000027 0.00067857 - layer.1.k_cache 0.00294478 5.91905230 - layer.1.v_cache 0.00000080 0.00280348 - layer.2.k_cache 0.00115317 2.09148227 - layer.2.v_cache 0.00000113 0.00428570 - layer.3.k_cache 0.00130758 2.45473300 - layer.3.v_cache 0.00000212 0.00730615 - layer.4.k_cache 0.00357902 6.50646835 - layer.4.v_cache 0.00000333 0.01213944 - layer.4.output 0.00014592 0.19145597 - ------------------------------------------------------------------------------------- - TOTAL 0.00245352 4.88747654 - (elements=5,103,616) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5103616 -Total Bytes 44516 -BPFP 0.0698 bits/point -EBPFP 0.1396 equivalent bits/point -MSE 4.887477 ----------------------- -------------------------------------------------------- -Time: 1.090s Load: 0.019s, Pack+Encode: 0.401s, Decode+Unpack: 0.670s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 356, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 356, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8875 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample65-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample65-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample66-layer4-item1.zst (74/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample66-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 342, 128) -Output shape: (1, 342, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) -> torch.Size([1, 1, 342, 1024]) - layer.4.output: torch.Size([1, 342, 4096]) -> torch.Size([1, 1, 342, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,572B, BPFP=0.0588 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,772B, BPFP=0.0633 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,784B, BPFP=0.0864 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,284B, BPFP=0.0750 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,000B, BPFP=0.0914 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,480B, BPFP=0.0795 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,196B, BPFP=0.0959 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,640B, BPFP=0.0832 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,644B, BPFP=0.0832 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,600B, BPFP=0.0822 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,100B, BPFP=0.0577 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.667s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 342, 128]) - layer.0.v_cache: torch.Size([1, 8, 342, 128]) - layer.1.k_cache: torch.Size([1, 8, 342, 128]) - layer.1.v_cache: torch.Size([1, 8, 342, 128]) - layer.2.k_cache: torch.Size([1, 8, 342, 128]) - layer.2.v_cache: torch.Size([1, 8, 342, 128]) - layer.3.k_cache: torch.Size([1, 8, 342, 128]) - layer.3.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.k_cache: torch.Size([1, 8, 342, 128]) - layer.4.v_cache: torch.Size([1, 8, 342, 128]) - layer.4.output: torch.Size([1, 342, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02260964 66.48145673 - layer.0.v_cache 0.00000027 0.00068136 - layer.1.k_cache 0.00288942 6.00888615 - layer.1.v_cache 0.00000084 0.00283637 - layer.2.k_cache 0.00116702 2.10081348 - layer.2.v_cache 0.00000122 0.00437632 - layer.3.k_cache 0.00131477 2.43738810 - layer.3.v_cache 0.00000216 0.00738730 - layer.4.k_cache 0.00350203 6.38614338 - layer.4.v_cache 0.00000356 0.01239867 - layer.4.output 0.00013713 0.19220595 - ------------------------------------------------------------------------------------- - TOTAL 0.00228853 6.01508512 - (elements=4,902,912) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4902912 -Total Bytes 45072 -BPFP 0.0735 bits/point -EBPFP 0.1471 equivalent bits/point -MSE 6.015085 ----------------------- -------------------------------------------------------- -Time: 1.096s Load: 0.018s, Pack+Encode: 0.411s, Decode+Unpack: 0.667s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 342, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 342, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 6.0151 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample66-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample66-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample67-layer4-item1.zst (75/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample67-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 360, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 360, 128) -Output shape: (1, 360, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.0.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.1.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.1.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.2.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.2.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.3.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.3.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.4.k_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.4.v_cache: torch.Size([1, 8, 360, 128]) -> torch.Size([1, 1, 360, 1024]) - layer.4.output: torch.Size([1, 360, 4096]) -> torch.Size([1, 1, 360, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,716B, BPFP=0.0589 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,916B, BPFP=0.0633 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,900B, BPFP=0.0846 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,408B, BPFP=0.0740 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,088B, BPFP=0.0887 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,540B, BPFP=0.0768 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,264B, BPFP=0.0925 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,796B, BPFP=0.0824 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,764B, BPFP=0.0817 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,668B, BPFP=0.0796 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,392B, BPFP=0.0564 -⌛️ [2/4] FRONTEND: Frontend time: 0.422s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 360, 128]) - layer.0.v_cache: torch.Size([1, 8, 360, 128]) - layer.1.k_cache: torch.Size([1, 8, 360, 128]) - layer.1.v_cache: torch.Size([1, 8, 360, 128]) - layer.2.k_cache: torch.Size([1, 8, 360, 128]) - layer.2.v_cache: torch.Size([1, 8, 360, 128]) - layer.3.k_cache: torch.Size([1, 8, 360, 128]) - layer.3.v_cache: torch.Size([1, 8, 360, 128]) - layer.4.k_cache: torch.Size([1, 8, 360, 128]) - layer.4.v_cache: torch.Size([1, 8, 360, 128]) - layer.4.output: torch.Size([1, 360, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.668s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 360, 128]) - layer.0.v_cache: torch.Size([1, 8, 360, 128]) - layer.1.k_cache: torch.Size([1, 8, 360, 128]) - layer.1.v_cache: torch.Size([1, 8, 360, 128]) - layer.2.k_cache: torch.Size([1, 8, 360, 128]) - layer.2.v_cache: torch.Size([1, 8, 360, 128]) - layer.3.k_cache: torch.Size([1, 8, 360, 128]) - layer.3.v_cache: torch.Size([1, 8, 360, 128]) - layer.4.k_cache: torch.Size([1, 8, 360, 128]) - layer.4.v_cache: torch.Size([1, 8, 360, 128]) - layer.4.output: torch.Size([1, 360, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02342221 45.93993056 - layer.0.v_cache 0.00000028 0.00067951 - layer.1.k_cache 0.00285409 5.77795342 - layer.1.v_cache 0.00000079 0.00282893 - layer.2.k_cache 0.00117169 2.12065447 - layer.2.v_cache 0.00000113 0.00429597 - layer.3.k_cache 0.00130395 2.44470011 - layer.3.v_cache 0.00000220 0.00751282 - layer.4.k_cache 0.00353751 6.31920844 - layer.4.v_cache 0.00000341 0.01260064 - layer.4.output 0.00014250 0.19089385 - ------------------------------------------------------------------------------------- - TOTAL 0.00234766 4.52813859 - (elements=5,160,960) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5160960 -Total Bytes 46452 -BPFP 0.0720 bits/point -EBPFP 0.1440 equivalent bits/point -MSE 4.528139 ----------------------- -------------------------------------------------------- -Time: 1.109s Load: 0.019s, Pack+Encode: 0.422s, Decode+Unpack: 0.668s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 360, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 360, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.5281 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample67-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample67-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample68-layer4-item1.zst (76/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample68-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 294, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 294, 128) -Output shape: (1, 294, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.0.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.1.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.1.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.2.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.2.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.3.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.3.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.4.k_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.4.v_cache: torch.Size([1, 8, 294, 128]) -> torch.Size([1, 1, 294, 1024]) - layer.4.output: torch.Size([1, 294, 4096]) -> torch.Size([1, 1, 294, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,252B, BPFP=0.0598 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,436B, BPFP=0.0647 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,240B, BPFP=0.0861 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,824B, BPFP=0.0750 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,384B, BPFP=0.0899 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,944B, BPFP=0.0782 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,548B, BPFP=0.0943 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,168B, BPFP=0.0842 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,116B, BPFP=0.0828 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,084B, BPFP=0.0820 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,000B, BPFP=0.0598 -⌛️ [2/4] FRONTEND: Frontend time: 0.357s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 294, 128]) - layer.0.v_cache: torch.Size([1, 8, 294, 128]) - layer.1.k_cache: torch.Size([1, 8, 294, 128]) - layer.1.v_cache: torch.Size([1, 8, 294, 128]) - layer.2.k_cache: torch.Size([1, 8, 294, 128]) - layer.2.v_cache: torch.Size([1, 8, 294, 128]) - layer.3.k_cache: torch.Size([1, 8, 294, 128]) - layer.3.v_cache: torch.Size([1, 8, 294, 128]) - layer.4.k_cache: torch.Size([1, 8, 294, 128]) - layer.4.v_cache: torch.Size([1, 8, 294, 128]) - layer.4.output: torch.Size([1, 294, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.569s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 294, 128]) - layer.0.v_cache: torch.Size([1, 8, 294, 128]) - layer.1.k_cache: torch.Size([1, 8, 294, 128]) - layer.1.v_cache: torch.Size([1, 8, 294, 128]) - layer.2.k_cache: torch.Size([1, 8, 294, 128]) - layer.2.v_cache: torch.Size([1, 8, 294, 128]) - layer.3.k_cache: torch.Size([1, 8, 294, 128]) - layer.3.v_cache: torch.Size([1, 8, 294, 128]) - layer.4.k_cache: torch.Size([1, 8, 294, 128]) - layer.4.v_cache: torch.Size([1, 8, 294, 128]) - layer.4.output: torch.Size([1, 294, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02439844 55.82249150 - layer.0.v_cache 0.00000027 0.00067232 - layer.1.k_cache 0.00289989 5.99103987 - layer.1.v_cache 0.00000083 0.00276762 - layer.2.k_cache 0.00119190 2.12993970 - layer.2.v_cache 0.00000115 0.00438791 - layer.3.k_cache 0.00131407 2.46323504 - layer.3.v_cache 0.00000220 0.00748108 - layer.4.k_cache 0.00351532 6.38108359 - layer.4.v_cache 0.00000333 0.01230047 - layer.4.output 0.00015801 0.20885304 - ------------------------------------------------------------------------------------- - TOTAL 0.00242567 5.26077223 - (elements=4,214,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4214784 -Total Bytes 38996 -BPFP 0.0740 bits/point -EBPFP 0.1480 equivalent bits/point -MSE 5.260772 ----------------------- -------------------------------------------------------- -Time: 0.942s Load: 0.016s, Pack+Encode: 0.357s, Decode+Unpack: 0.569s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 294, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 294, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.2608 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample68-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample68-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample69-layer4-item1.zst (77/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample69-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 338, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 338, 128) -Output shape: (1, 338, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.0.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.1.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.1.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.2.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.2.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.3.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.3.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.4.k_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.4.v_cache: torch.Size([1, 8, 338, 128]) -> torch.Size([1, 1, 338, 1024]) - layer.4.output: torch.Size([1, 338, 4096]) -> torch.Size([1, 1, 338, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,520B, BPFP=0.0582 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,776B, BPFP=0.0642 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,792B, BPFP=0.0876 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,256B, BPFP=0.0753 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,000B, BPFP=0.0925 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,436B, BPFP=0.0794 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,224B, BPFP=0.0976 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,696B, BPFP=0.0854 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,664B, BPFP=0.0847 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,604B, BPFP=0.0833 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,028B, BPFP=0.0579 -⌛️ [2/4] FRONTEND: Frontend time: 0.407s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 338, 128]) - layer.0.v_cache: torch.Size([1, 8, 338, 128]) - layer.1.k_cache: torch.Size([1, 8, 338, 128]) - layer.1.v_cache: torch.Size([1, 8, 338, 128]) - layer.2.k_cache: torch.Size([1, 8, 338, 128]) - layer.2.v_cache: torch.Size([1, 8, 338, 128]) - layer.3.k_cache: torch.Size([1, 8, 338, 128]) - layer.3.v_cache: torch.Size([1, 8, 338, 128]) - layer.4.k_cache: torch.Size([1, 8, 338, 128]) - layer.4.v_cache: torch.Size([1, 8, 338, 128]) - layer.4.output: torch.Size([1, 338, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.671s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 338, 128]) - layer.0.v_cache: torch.Size([1, 8, 338, 128]) - layer.1.k_cache: torch.Size([1, 8, 338, 128]) - layer.1.v_cache: torch.Size([1, 8, 338, 128]) - layer.2.k_cache: torch.Size([1, 8, 338, 128]) - layer.2.v_cache: torch.Size([1, 8, 338, 128]) - layer.3.k_cache: torch.Size([1, 8, 338, 128]) - layer.3.v_cache: torch.Size([1, 8, 338, 128]) - layer.4.k_cache: torch.Size([1, 8, 338, 128]) - layer.4.v_cache: torch.Size([1, 8, 338, 128]) - layer.4.output: torch.Size([1, 338, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02370679 47.59749156 - layer.0.v_cache 0.00000026 0.00066429 - layer.1.k_cache 0.00288373 5.82016226 - layer.1.v_cache 0.00000083 0.00278456 - layer.2.k_cache 0.00117924 2.13053361 - layer.2.v_cache 0.00000113 0.00433994 - layer.3.k_cache 0.00131114 2.44598967 - layer.3.v_cache 0.00000214 0.00750327 - layer.4.k_cache 0.00355340 6.37672560 - layer.4.v_cache 0.00000355 0.01242336 - layer.4.output 0.00015048 0.19866065 - ------------------------------------------------------------------------------------- - TOTAL 0.00237458 4.65666148 - (elements=4,845,568) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4845568 -Total Bytes 44996 -BPFP 0.0743 bits/point -EBPFP 0.1486 equivalent bits/point -MSE 4.656661 ----------------------- -------------------------------------------------------- -Time: 1.095s Load: 0.017s, Pack+Encode: 0.407s, Decode+Unpack: 0.671s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 338, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 338, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.6567 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample69-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample69-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample7-layer4-item1.zst (78/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample7-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 340, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 340, 128) -Output shape: (1, 340, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) -> torch.Size([1, 1, 340, 1024]) - layer.4.output: torch.Size([1, 340, 4096]) -> torch.Size([1, 1, 340, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,532B, BPFP=0.0582 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,832B, BPFP=0.0651 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,784B, BPFP=0.0869 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,308B, BPFP=0.0760 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,040B, BPFP=0.0928 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,424B, BPFP=0.0787 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,192B, BPFP=0.0963 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,724B, BPFP=0.0856 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,624B, BPFP=0.0833 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,640B, BPFP=0.0836 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,400B, BPFP=0.0597 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.output: torch.Size([1, 340, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.669s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 340, 128]) - layer.0.v_cache: torch.Size([1, 8, 340, 128]) - layer.1.k_cache: torch.Size([1, 8, 340, 128]) - layer.1.v_cache: torch.Size([1, 8, 340, 128]) - layer.2.k_cache: torch.Size([1, 8, 340, 128]) - layer.2.v_cache: torch.Size([1, 8, 340, 128]) - layer.3.k_cache: torch.Size([1, 8, 340, 128]) - layer.3.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.k_cache: torch.Size([1, 8, 340, 128]) - layer.4.v_cache: torch.Size([1, 8, 340, 128]) - layer.4.output: torch.Size([1, 340, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02367318 54.88184743 - layer.0.v_cache 0.00000026 0.00067196 - layer.1.k_cache 0.00294777 5.85814532 - layer.1.v_cache 0.00000081 0.00285031 - layer.2.k_cache 0.00121079 2.11927490 - layer.2.v_cache 0.00000125 0.00438163 - layer.3.k_cache 0.00131075 2.43376429 - layer.3.v_cache 0.00000218 0.00754033 - layer.4.k_cache 0.00357831 6.40866125 - layer.4.v_cache 0.00000337 0.01264371 - layer.4.output 0.00014951 0.20503309 - ------------------------------------------------------------------------------------- - TOTAL 0.00238048 5.18213668 - (elements=4,874,240) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4874240 -Total Bytes 45500 -BPFP 0.0747 bits/point -EBPFP 0.1494 equivalent bits/point -MSE 5.182137 ----------------------- -------------------------------------------------------- -Time: 1.099s Load: 0.019s, Pack+Encode: 0.411s, Decode+Unpack: 0.669s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 340, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 340, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1821 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample7-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample7-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample70-layer4-item1.zst (79/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample70-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 320, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.016s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 320, 128) -Output shape: (1, 320, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) -> torch.Size([1, 1, 320, 1024]) - layer.4.output: torch.Size([1, 320, 4096]) -> torch.Size([1, 1, 320, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,932B, BPFP=0.0472 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,040B, BPFP=0.0498 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,804B, BPFP=0.0685 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,320B, BPFP=0.0566 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,064B, BPFP=0.0748 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,500B, BPFP=0.0610 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,196B, BPFP=0.0780 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,688B, BPFP=0.0656 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,692B, BPFP=0.0657 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,588B, BPFP=0.0632 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 7,148B, BPFP=0.0436 -⌛️ [2/4] FRONTEND: Frontend time: 0.356s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.output: torch.Size([1, 320, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.569s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 320, 128]) - layer.0.v_cache: torch.Size([1, 8, 320, 128]) - layer.1.k_cache: torch.Size([1, 8, 320, 128]) - layer.1.v_cache: torch.Size([1, 8, 320, 128]) - layer.2.k_cache: torch.Size([1, 8, 320, 128]) - layer.2.v_cache: torch.Size([1, 8, 320, 128]) - layer.3.k_cache: torch.Size([1, 8, 320, 128]) - layer.3.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.k_cache: torch.Size([1, 8, 320, 128]) - layer.4.v_cache: torch.Size([1, 8, 320, 128]) - layer.4.output: torch.Size([1, 320, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02419060 34.59775391 - layer.0.v_cache 0.00000027 0.00065471 - layer.1.k_cache 0.00288080 5.45278168 - layer.1.v_cache 0.00000084 0.00281212 - layer.2.k_cache 0.00117023 2.00977631 - layer.2.v_cache 0.00000114 0.00430736 - layer.3.k_cache 0.00131996 2.33582783 - layer.3.v_cache 0.00000215 0.00726224 - layer.4.k_cache 0.00348745 6.10306625 - layer.4.v_cache 0.00000346 0.01210943 - layer.4.output 0.00014756 0.19189711 - ------------------------------------------------------------------------------------- - TOTAL 0.00240337 3.66385287 - (elements=4,587,520) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4587520 -Total Bytes 32972 -BPFP 0.0575 bits/point -EBPFP 0.1150 equivalent bits/point -MSE 3.663853 ----------------------- -------------------------------------------------------- -Time: 0.941s Load: 0.016s, Pack+Encode: 0.356s, Decode+Unpack: 0.569s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 320, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 320, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 3.6639 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample70-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample70-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample71-layer4-item1.zst (80/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample71-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 328, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 328, 128) -Output shape: (1, 328, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.output: torch.Size([1, 328, 4096]) -> torch.Size([1, 1, 328, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,648B, BPFP=0.0631 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,772B, BPFP=0.0660 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,764B, BPFP=0.0897 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,308B, BPFP=0.0788 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,040B, BPFP=0.0962 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,416B, BPFP=0.0814 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,200B, BPFP=0.1000 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,640B, BPFP=0.0867 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,608B, BPFP=0.0859 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,592B, BPFP=0.0856 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,508B, BPFP=0.0626 -⌛️ [2/4] FRONTEND: Frontend time: 0.420s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.output: torch.Size([1, 328, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.668s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.output: torch.Size([1, 328, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02411815 47.65206090 - layer.0.v_cache 0.00000026 0.00067437 - layer.1.k_cache 0.00293126 5.87598252 - layer.1.v_cache 0.00000079 0.00276750 - layer.2.k_cache 0.00115822 2.07161936 - layer.2.v_cache 0.00000116 0.00425343 - layer.3.k_cache 0.00132993 2.41625195 - layer.3.v_cache 0.00000214 0.00720070 - layer.4.k_cache 0.00358323 6.49327869 - layer.4.v_cache 0.00000336 0.01202088 - layer.4.output 0.00014415 0.19661057 - ------------------------------------------------------------------------------------- - TOTAL 0.00240751 4.66589661 - (elements=4,702,208) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4702208 -Total Bytes 45496 -BPFP 0.0774 bits/point -EBPFP 0.1548 equivalent bits/point -MSE 4.665897 ----------------------- -------------------------------------------------------- -Time: 1.107s Load: 0.018s, Pack+Encode: 0.420s, Decode+Unpack: 0.668s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 328, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.6659 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample71-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample71-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample72-layer4-item1.zst (81/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample72-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 328, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 328, 128) -Output shape: (1, 328, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) -> torch.Size([1, 1, 328, 1024]) - layer.4.output: torch.Size([1, 328, 4096]) -> torch.Size([1, 1, 328, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,660B, BPFP=0.0634 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,780B, BPFP=0.0662 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,744B, BPFP=0.0892 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,260B, BPFP=0.0776 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,076B, BPFP=0.0971 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,460B, BPFP=0.0824 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,184B, BPFP=0.0997 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,664B, BPFP=0.0873 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,600B, BPFP=0.0857 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,584B, BPFP=0.0854 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,540B, BPFP=0.0628 -⌛️ [2/4] FRONTEND: Frontend time: 0.414s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.output: torch.Size([1, 328, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.670s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 328, 128]) - layer.0.v_cache: torch.Size([1, 8, 328, 128]) - layer.1.k_cache: torch.Size([1, 8, 328, 128]) - layer.1.v_cache: torch.Size([1, 8, 328, 128]) - layer.2.k_cache: torch.Size([1, 8, 328, 128]) - layer.2.v_cache: torch.Size([1, 8, 328, 128]) - layer.3.k_cache: torch.Size([1, 8, 328, 128]) - layer.3.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.k_cache: torch.Size([1, 8, 328, 128]) - layer.4.v_cache: torch.Size([1, 8, 328, 128]) - layer.4.output: torch.Size([1, 328, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02582944 48.27296113 - layer.0.v_cache 0.00000026 0.00066290 - layer.1.k_cache 0.00293683 5.77150819 - layer.1.v_cache 0.00000081 0.00280032 - layer.2.k_cache 0.00118071 2.10415128 - layer.2.v_cache 0.00000117 0.00439594 - layer.3.k_cache 0.00133336 2.44277675 - layer.3.v_cache 0.00000214 0.00727953 - layer.4.k_cache 0.00361319 6.42286347 - layer.4.v_cache 0.00000321 0.01184127 - layer.4.output 0.00015382 0.20288963 - ------------------------------------------------------------------------------------- - TOTAL 0.00253689 4.70377138 - (elements=4,702,208) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4702208 -Total Bytes 45552 -BPFP 0.0775 bits/point -EBPFP 0.1550 equivalent bits/point -MSE 4.703771 ----------------------- -------------------------------------------------------- -Time: 1.101s Load: 0.017s, Pack+Encode: 0.414s, Decode+Unpack: 0.670s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 328, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 328, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7038 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample72-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample72-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample73-layer4-item1.zst (82/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample73-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 352, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 352, 128) -Output shape: (1, 352, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.0.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.1.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.1.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.2.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.2.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.3.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.3.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.4.k_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.4.v_cache: torch.Size([1, 8, 352, 128]) -> torch.Size([1, 1, 352, 1024]) - layer.4.output: torch.Size([1, 352, 4096]) -> torch.Size([1, 1, 352, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,512B, BPFP=0.0558 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,700B, BPFP=0.0599 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,652B, BPFP=0.0811 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,072B, BPFP=0.0682 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,900B, BPFP=0.0866 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,280B, BPFP=0.0728 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,064B, BPFP=0.0902 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,548B, BPFP=0.0787 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,540B, BPFP=0.0786 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,488B, BPFP=0.0774 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,856B, BPFP=0.0547 -⌛️ [2/4] FRONTEND: Frontend time: 0.418s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 352, 128]) - layer.0.v_cache: torch.Size([1, 8, 352, 128]) - layer.1.k_cache: torch.Size([1, 8, 352, 128]) - layer.1.v_cache: torch.Size([1, 8, 352, 128]) - layer.2.k_cache: torch.Size([1, 8, 352, 128]) - layer.2.v_cache: torch.Size([1, 8, 352, 128]) - layer.3.k_cache: torch.Size([1, 8, 352, 128]) - layer.3.v_cache: torch.Size([1, 8, 352, 128]) - layer.4.k_cache: torch.Size([1, 8, 352, 128]) - layer.4.v_cache: torch.Size([1, 8, 352, 128]) - layer.4.output: torch.Size([1, 352, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.669s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 352, 128]) - layer.0.v_cache: torch.Size([1, 8, 352, 128]) - layer.1.k_cache: torch.Size([1, 8, 352, 128]) - layer.1.v_cache: torch.Size([1, 8, 352, 128]) - layer.2.k_cache: torch.Size([1, 8, 352, 128]) - layer.2.v_cache: torch.Size([1, 8, 352, 128]) - layer.3.k_cache: torch.Size([1, 8, 352, 128]) - layer.3.v_cache: torch.Size([1, 8, 352, 128]) - layer.4.k_cache: torch.Size([1, 8, 352, 128]) - layer.4.v_cache: torch.Size([1, 8, 352, 128]) - layer.4.output: torch.Size([1, 352, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02395736 62.38505970 - layer.0.v_cache 0.00000026 0.00067278 - layer.1.k_cache 0.00288394 6.02020957 - layer.1.v_cache 0.00000077 0.00277226 - layer.2.k_cache 0.00114480 2.07586895 - layer.2.v_cache 0.00000114 0.00434108 - layer.3.k_cache 0.00130173 2.43437576 - layer.3.v_cache 0.00000215 0.00742137 - layer.4.k_cache 0.00352615 6.33236764 - layer.4.v_cache 0.00000336 0.01247274 - layer.4.output 0.00014232 0.19189135 - ------------------------------------------------------------------------------------- - TOTAL 0.00238507 5.71736623 - (elements=5,046,272) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5046272 -Total Bytes 43612 -BPFP 0.0691 bits/point -EBPFP 0.1383 equivalent bits/point -MSE 5.717366 ----------------------- -------------------------------------------------------- -Time: 1.105s Load: 0.017s, Pack+Encode: 0.418s, Decode+Unpack: 0.669s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 352, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 352, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.7174 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample73-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample73-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample74-layer4-item1.zst (83/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample74-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 323, 128) -Output shape: (1, 323, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.output: torch.Size([1, 323, 4096]) -> torch.Size([1, 1, 323, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,576B, BPFP=0.0623 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,724B, BPFP=0.0659 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,688B, BPFP=0.0892 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,108B, BPFP=0.0752 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,972B, BPFP=0.0961 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,284B, BPFP=0.0794 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,124B, BPFP=0.0997 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,520B, BPFP=0.0851 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,524B, BPFP=0.0852 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,444B, BPFP=0.0833 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,508B, BPFP=0.0635 -⌛️ [2/4] FRONTEND: Frontend time: 0.417s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.665s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02435790 65.23384288 - layer.0.v_cache 0.00000027 0.00068227 - layer.1.k_cache 0.00290123 5.84927491 - layer.1.v_cache 0.00000082 0.00280683 - layer.2.k_cache 0.00116542 2.12720955 - layer.2.v_cache 0.00000118 0.00445764 - layer.3.k_cache 0.00132756 2.45836987 - layer.3.v_cache 0.00000224 0.00766112 - layer.4.k_cache 0.00351721 6.46464420 - layer.4.v_cache 0.00000335 0.01261965 - layer.4.output 0.00015064 0.20859920 - ------------------------------------------------------------------------------------- - TOTAL 0.00241998 5.92828326 - (elements=4,630,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4630528 -Total Bytes 44472 -BPFP 0.0768 bits/point -EBPFP 0.1537 equivalent bits/point -MSE 5.928283 ----------------------- -------------------------------------------------------- -Time: 1.101s Load: 0.018s, Pack+Encode: 0.417s, Decode+Unpack: 0.665s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.9283 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample74-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample74-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample75-layer4-item1.zst (84/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample75-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 327, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 327, 128) -Output shape: (1, 327, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) -> torch.Size([1, 1, 327, 1024]) - layer.4.output: torch.Size([1, 327, 4096]) -> torch.Size([1, 1, 327, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,620B, BPFP=0.0626 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,816B, BPFP=0.0673 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,756B, BPFP=0.0897 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,288B, BPFP=0.0786 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,044B, BPFP=0.0966 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,500B, BPFP=0.0836 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,184B, BPFP=0.1000 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,684B, BPFP=0.0880 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,600B, BPFP=0.0860 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,644B, BPFP=0.0871 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,752B, BPFP=0.0642 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.output: torch.Size([1, 327, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.670s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 327, 128]) - layer.0.v_cache: torch.Size([1, 8, 327, 128]) - layer.1.k_cache: torch.Size([1, 8, 327, 128]) - layer.1.v_cache: torch.Size([1, 8, 327, 128]) - layer.2.k_cache: torch.Size([1, 8, 327, 128]) - layer.2.v_cache: torch.Size([1, 8, 327, 128]) - layer.3.k_cache: torch.Size([1, 8, 327, 128]) - layer.3.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.k_cache: torch.Size([1, 8, 327, 128]) - layer.4.v_cache: torch.Size([1, 8, 327, 128]) - layer.4.output: torch.Size([1, 327, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02420299 54.01336726 - layer.0.v_cache 0.00000027 0.00066490 - layer.1.k_cache 0.00291079 5.84935105 - layer.1.v_cache 0.00000081 0.00283593 - layer.2.k_cache 0.00118771 2.09293620 - layer.2.v_cache 0.00000120 0.00444091 - layer.3.k_cache 0.00132021 2.44100047 - layer.3.v_cache 0.00000219 0.00754224 - layer.4.k_cache 0.00351552 6.51254674 - layer.4.v_cache 0.00000327 0.01224386 - layer.4.output 0.00014986 0.20546797 - ------------------------------------------------------------------------------------- - TOTAL 0.00241031 5.12562867 - (elements=4,687,872) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4687872 -Total Bytes 45888 -BPFP 0.0783 bits/point -EBPFP 0.1566 equivalent bits/point -MSE 5.125629 ----------------------- -------------------------------------------------------- -Time: 1.098s Load: 0.018s, Pack+Encode: 0.411s, Decode+Unpack: 0.670s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 327, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 327, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1256 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample75-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample75-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample76-layer4-item1.zst (85/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample76-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 330, 128) -Output shape: (1, 330, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) -> torch.Size([1, 1, 330, 1024]) - layer.4.output: torch.Size([1, 330, 4096]) -> torch.Size([1, 1, 330, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,608B, BPFP=0.0617 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,764B, BPFP=0.0654 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,764B, BPFP=0.0891 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,312B, BPFP=0.0784 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,088B, BPFP=0.0968 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,524B, BPFP=0.0834 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,196B, BPFP=0.0993 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,768B, BPFP=0.0892 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,620B, BPFP=0.0857 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,564B, BPFP=0.0844 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,400B, BPFP=0.0616 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.657s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 330, 128]) - layer.0.v_cache: torch.Size([1, 8, 330, 128]) - layer.1.k_cache: torch.Size([1, 8, 330, 128]) - layer.1.v_cache: torch.Size([1, 8, 330, 128]) - layer.2.k_cache: torch.Size([1, 8, 330, 128]) - layer.2.v_cache: torch.Size([1, 8, 330, 128]) - layer.3.k_cache: torch.Size([1, 8, 330, 128]) - layer.3.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.k_cache: torch.Size([1, 8, 330, 128]) - layer.4.v_cache: torch.Size([1, 8, 330, 128]) - layer.4.output: torch.Size([1, 330, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02537568 52.82220052 - layer.0.v_cache 0.00000027 0.00066957 - layer.1.k_cache 0.00294190 5.77373824 - layer.1.v_cache 0.00000076 0.00267993 - layer.2.k_cache 0.00116180 2.08954560 - layer.2.v_cache 0.00000113 0.00422102 - layer.3.k_cache 0.00131353 2.42714640 - layer.3.v_cache 0.00000217 0.00720224 - layer.4.k_cache 0.00362690 6.43585982 - layer.4.v_cache 0.00000322 0.01183365 - layer.4.output 0.00015895 0.20954130 - ------------------------------------------------------------------------------------- - TOTAL 0.00250451 5.02951873 - (elements=4,730,880) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4730880 -Total Bytes 45608 -BPFP 0.0771 bits/point -EBPFP 0.1542 equivalent bits/point -MSE 5.029519 ----------------------- -------------------------------------------------------- -Time: 1.080s Load: 0.017s, Pack+Encode: 0.406s, Decode+Unpack: 0.657s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 330, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 330, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.0295 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample76-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample76-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample77-layer4-item1.zst (86/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample77-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,652B, BPFP=0.0630 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,780B, BPFP=0.0660 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,788B, BPFP=0.0900 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,320B, BPFP=0.0788 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,060B, BPFP=0.0964 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,468B, BPFP=0.0824 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,204B, BPFP=0.0998 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,716B, BPFP=0.0882 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,636B, BPFP=0.0863 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,552B, BPFP=0.0843 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,300B, BPFP=0.0611 -⌛️ [2/4] FRONTEND: Frontend time: 0.405s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.659s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02321163 50.35166342 - layer.0.v_cache 0.00000026 0.00066553 - layer.1.k_cache 0.00292509 5.74744803 - layer.1.v_cache 0.00000078 0.00277227 - layer.2.k_cache 0.00118155 2.12345798 - layer.2.v_cache 0.00000112 0.00421482 - layer.3.k_cache 0.00133849 2.44817614 - layer.3.v_cache 0.00000208 0.00704468 - layer.4.k_cache 0.00360187 6.47490620 - layer.4.v_cache 0.00000304 0.01172412 - layer.4.output 0.00015018 0.19674474 - ------------------------------------------------------------------------------------- - TOTAL 0.00234762 4.85421801 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 45476 -BPFP 0.0771 bits/point -EBPFP 0.1543 equivalent bits/point -MSE 4.854218 ----------------------- -------------------------------------------------------- -Time: 1.082s Load: 0.018s, Pack+Encode: 0.405s, Decode+Unpack: 0.659s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.8542 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample77-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample77-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample78-layer4-item1.zst (87/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample78-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 333, 128) -Output shape: (1, 333, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.output: torch.Size([1, 333, 4096]) -> torch.Size([1, 1, 333, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,588B, BPFP=0.0607 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,776B, BPFP=0.0651 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,804B, BPFP=0.0892 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,264B, BPFP=0.0766 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,032B, BPFP=0.0946 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,452B, BPFP=0.0810 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,204B, BPFP=0.0986 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,664B, BPFP=0.0860 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,612B, BPFP=0.0847 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,584B, BPFP=0.0841 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,056B, BPFP=0.0590 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.683s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02306361 51.49882109 - layer.0.v_cache 0.00000026 0.00066282 - layer.1.k_cache 0.00286224 5.78251029 - layer.1.v_cache 0.00000078 0.00272742 - layer.2.k_cache 0.00116467 2.10043967 - layer.2.v_cache 0.00000115 0.00430271 - layer.3.k_cache 0.00135416 2.46252020 - layer.3.v_cache 0.00000234 0.00751490 - layer.4.k_cache 0.00356469 6.39803793 - layer.4.v_cache 0.00000325 0.01217190 - layer.4.output 0.00015110 0.20670041 - ------------------------------------------------------------------------------------- - TOTAL 0.00233011 4.93546504 - (elements=4,773,888) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4773888 -Total Bytes 45036 -BPFP 0.0755 bits/point -EBPFP 0.1509 equivalent bits/point -MSE 4.935465 ----------------------- -------------------------------------------------------- -Time: 1.113s Load: 0.019s, Pack+Encode: 0.412s, Decode+Unpack: 0.683s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.9355 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample78-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample78-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample79-layer4-item1.zst (88/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample79-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 304, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.017s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 304, 128) -Output shape: (1, 304, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.0.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.1.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.1.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.2.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.2.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.3.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.3.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.4.k_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.4.v_cache: torch.Size([1, 8, 304, 128]) -> torch.Size([1, 1, 304, 1024]) - layer.4.output: torch.Size([1, 304, 4096]) -> torch.Size([1, 1, 304, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,408B, BPFP=0.0619 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,560B, BPFP=0.0658 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,348B, BPFP=0.0860 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,900B, BPFP=0.0745 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,504B, BPFP=0.0900 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,040B, BPFP=0.0781 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,660B, BPFP=0.0941 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,232B, BPFP=0.0831 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,228B, BPFP=0.0830 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,176B, BPFP=0.0816 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 8,876B, BPFP=0.0570 -⌛️ [2/4] FRONTEND: Frontend time: 0.358s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 304, 128]) - layer.0.v_cache: torch.Size([1, 8, 304, 128]) - layer.1.k_cache: torch.Size([1, 8, 304, 128]) - layer.1.v_cache: torch.Size([1, 8, 304, 128]) - layer.2.k_cache: torch.Size([1, 8, 304, 128]) - layer.2.v_cache: torch.Size([1, 8, 304, 128]) - layer.3.k_cache: torch.Size([1, 8, 304, 128]) - layer.3.v_cache: torch.Size([1, 8, 304, 128]) - layer.4.k_cache: torch.Size([1, 8, 304, 128]) - layer.4.v_cache: torch.Size([1, 8, 304, 128]) - layer.4.output: torch.Size([1, 304, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.572s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 304, 128]) - layer.0.v_cache: torch.Size([1, 8, 304, 128]) - layer.1.k_cache: torch.Size([1, 8, 304, 128]) - layer.1.v_cache: torch.Size([1, 8, 304, 128]) - layer.2.k_cache: torch.Size([1, 8, 304, 128]) - layer.2.v_cache: torch.Size([1, 8, 304, 128]) - layer.3.k_cache: torch.Size([1, 8, 304, 128]) - layer.3.v_cache: torch.Size([1, 8, 304, 128]) - layer.4.k_cache: torch.Size([1, 8, 304, 128]) - layer.4.v_cache: torch.Size([1, 8, 304, 128]) - layer.4.output: torch.Size([1, 304, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02416075 56.95989669 - layer.0.v_cache 0.00000027 0.00065617 - layer.1.k_cache 0.00290015 5.74881945 - layer.1.v_cache 0.00000081 0.00277909 - layer.2.k_cache 0.00116609 2.12507589 - layer.2.v_cache 0.00000115 0.00435322 - layer.3.k_cache 0.00131777 2.43825129 - layer.3.v_cache 0.00000216 0.00741321 - layer.4.k_cache 0.00349415 6.26297118 - layer.4.v_cache 0.00000336 0.01231695 - layer.4.output 0.00015143 0.20285777 - ------------------------------------------------------------------------------------- - TOTAL 0.00240374 5.31242602 - (elements=4,358,144) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4358144 -Total Bytes 39932 -BPFP 0.0733 bits/point -EBPFP 0.1466 equivalent bits/point -MSE 5.312426 ----------------------- -------------------------------------------------------- -Time: 0.947s Load: 0.017s, Pack+Encode: 0.358s, Decode+Unpack: 0.572s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 304, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 304, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.3124 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample79-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample79-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample8-layer4-item1.zst (89/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample8-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 351, 128) -Output shape: (1, 351, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) -> torch.Size([1, 1, 351, 1024]) - layer.4.output: torch.Size([1, 351, 4096]) -> torch.Size([1, 1, 351, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,496B, BPFP=0.0556 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,740B, BPFP=0.0610 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,648B, BPFP=0.0812 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,152B, BPFP=0.0702 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,916B, BPFP=0.0872 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,328B, BPFP=0.0741 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,068B, BPFP=0.0905 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,584B, BPFP=0.0798 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,524B, BPFP=0.0784 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,520B, BPFP=0.0783 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,160B, BPFP=0.0565 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.664s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 351, 128]) - layer.0.v_cache: torch.Size([1, 8, 351, 128]) - layer.1.k_cache: torch.Size([1, 8, 351, 128]) - layer.1.v_cache: torch.Size([1, 8, 351, 128]) - layer.2.k_cache: torch.Size([1, 8, 351, 128]) - layer.2.v_cache: torch.Size([1, 8, 351, 128]) - layer.3.k_cache: torch.Size([1, 8, 351, 128]) - layer.3.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.k_cache: torch.Size([1, 8, 351, 128]) - layer.4.v_cache: torch.Size([1, 8, 351, 128]) - layer.4.output: torch.Size([1, 351, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02287757 49.14805912 - layer.0.v_cache 0.00000026 0.00065132 - layer.1.k_cache 0.00288319 5.61828683 - layer.1.v_cache 0.00000079 0.00270289 - layer.2.k_cache 0.00116994 2.07320950 - layer.2.v_cache 0.00000119 0.00438203 - layer.3.k_cache 0.00131288 2.41393921 - layer.3.v_cache 0.00000224 0.00756804 - layer.4.k_cache 0.00358885 6.51612928 - layer.4.v_cache 0.00000339 0.01244306 - layer.4.output 0.00014945 0.19886015 - ------------------------------------------------------------------------------------- - TOTAL 0.00231701 4.75662942 - (elements=5,031,936) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5031936 -Total Bytes 44136 -BPFP 0.0702 bits/point -EBPFP 0.1403 equivalent bits/point -MSE 4.756629 ----------------------- -------------------------------------------------------- -Time: 1.096s Load: 0.019s, Pack+Encode: 0.412s, Decode+Unpack: 0.664s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 351, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 351, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.7566 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample8-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample8-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample80-layer4-item1.zst (90/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample80-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 333, 128) -Output shape: (1, 333, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) -> torch.Size([1, 1, 333, 1024]) - layer.4.output: torch.Size([1, 333, 4096]) -> torch.Size([1, 1, 333, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,584B, BPFP=0.0606 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,756B, BPFP=0.0647 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,828B, BPFP=0.0898 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,248B, BPFP=0.0762 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,008B, BPFP=0.0940 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,476B, BPFP=0.0816 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,188B, BPFP=0.0983 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,664B, BPFP=0.0860 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,600B, BPFP=0.0845 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,588B, BPFP=0.0842 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,116B, BPFP=0.0593 -⌛️ [2/4] FRONTEND: Frontend time: 0.415s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.666s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 333, 128]) - layer.0.v_cache: torch.Size([1, 8, 333, 128]) - layer.1.k_cache: torch.Size([1, 8, 333, 128]) - layer.1.v_cache: torch.Size([1, 8, 333, 128]) - layer.2.k_cache: torch.Size([1, 8, 333, 128]) - layer.2.v_cache: torch.Size([1, 8, 333, 128]) - layer.3.k_cache: torch.Size([1, 8, 333, 128]) - layer.3.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.k_cache: torch.Size([1, 8, 333, 128]) - layer.4.v_cache: torch.Size([1, 8, 333, 128]) - layer.4.output: torch.Size([1, 333, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02328842 53.14483624 - layer.0.v_cache 0.00000026 0.00065375 - layer.1.k_cache 0.00290993 5.92294101 - layer.1.v_cache 0.00000078 0.00270376 - layer.2.k_cache 0.00115179 2.10987116 - layer.2.v_cache 0.00000114 0.00430729 - layer.3.k_cache 0.00134347 2.46105865 - layer.3.v_cache 0.00000212 0.00733120 - layer.4.k_cache 0.00362086 6.54908546 - layer.4.v_cache 0.00000365 0.01197518 - layer.4.output 0.00013924 0.19714456 - ------------------------------------------------------------------------------------- - TOTAL 0.00234853 5.07166728 - (elements=4,773,888) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4773888 -Total Bytes 45056 -BPFP 0.0755 bits/point -EBPFP 0.1510 equivalent bits/point -MSE 5.071667 ----------------------- -------------------------------------------------------- -Time: 1.100s Load: 0.019s, Pack+Encode: 0.415s, Decode+Unpack: 0.666s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 333, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 333, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.0717 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample80-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample80-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample81-layer4-item1.zst (91/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample81-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 319, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 319, 128) -Output shape: (1, 319, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) -> torch.Size([1, 1, 319, 1024]) - layer.4.output: torch.Size([1, 319, 4096]) -> torch.Size([1, 1, 319, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 1,956B, BPFP=0.0479 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,052B, BPFP=0.0503 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 2,828B, BPFP=0.0693 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 2,388B, BPFP=0.0585 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,092B, BPFP=0.0757 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 2,524B, BPFP=0.0618 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 3,216B, BPFP=0.0788 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 2,712B, BPFP=0.0664 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 2,712B, BPFP=0.0664 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 2,644B, BPFP=0.0648 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 7,552B, BPFP=0.0462 -⌛️ [2/4] FRONTEND: Frontend time: 0.357s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.output: torch.Size([1, 319, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.562s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 319, 128]) - layer.0.v_cache: torch.Size([1, 8, 319, 128]) - layer.1.k_cache: torch.Size([1, 8, 319, 128]) - layer.1.v_cache: torch.Size([1, 8, 319, 128]) - layer.2.k_cache: torch.Size([1, 8, 319, 128]) - layer.2.v_cache: torch.Size([1, 8, 319, 128]) - layer.3.k_cache: torch.Size([1, 8, 319, 128]) - layer.3.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.k_cache: torch.Size([1, 8, 319, 128]) - layer.4.v_cache: torch.Size([1, 8, 319, 128]) - layer.4.output: torch.Size([1, 319, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02501136 39.52199868 - layer.0.v_cache 0.00000027 0.00065959 - layer.1.k_cache 0.00284013 5.53299304 - layer.1.v_cache 0.00000084 0.00281092 - layer.2.k_cache 0.00117936 2.07614356 - layer.2.v_cache 0.00000120 0.00451971 - layer.3.k_cache 0.00130692 2.37181775 - layer.3.v_cache 0.00000226 0.00759101 - layer.4.k_cache 0.00361670 6.28488685 - layer.4.v_cache 0.00000371 0.01278922 - layer.4.output 0.00016871 0.20612739 - ------------------------------------------------------------------------------------- - TOTAL 0.00247411 4.04576571 - (elements=4,573,184) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4573184 -Total Bytes 33676 -BPFP 0.0589 bits/point -EBPFP 0.1178 equivalent bits/point -MSE 4.045766 ----------------------- -------------------------------------------------------- -Time: 0.937s Load: 0.018s, Pack+Encode: 0.357s, Decode+Unpack: 0.562s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 319, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 319, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.0458 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample81-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample81-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample82-layer4-item1.zst (92/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample82-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 353, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.020s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 353, 128) -Output shape: (1, 353, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) -> torch.Size([1, 1, 353, 1024]) - layer.4.output: torch.Size([1, 353, 4096]) -> torch.Size([1, 1, 353, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,504B, BPFP=0.0554 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,756B, BPFP=0.0610 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,656B, BPFP=0.0809 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,176B, BPFP=0.0703 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,904B, BPFP=0.0864 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,320B, BPFP=0.0735 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,076B, BPFP=0.0902 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,596B, BPFP=0.0796 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,548B, BPFP=0.0785 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,512B, BPFP=0.0777 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,220B, BPFP=0.0565 -⌛️ [2/4] FRONTEND: Frontend time: 0.407s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.output: torch.Size([1, 353, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.659s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 353, 128]) - layer.0.v_cache: torch.Size([1, 8, 353, 128]) - layer.1.k_cache: torch.Size([1, 8, 353, 128]) - layer.1.v_cache: torch.Size([1, 8, 353, 128]) - layer.2.k_cache: torch.Size([1, 8, 353, 128]) - layer.2.v_cache: torch.Size([1, 8, 353, 128]) - layer.3.k_cache: torch.Size([1, 8, 353, 128]) - layer.3.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.k_cache: torch.Size([1, 8, 353, 128]) - layer.4.v_cache: torch.Size([1, 8, 353, 128]) - layer.4.output: torch.Size([1, 353, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02381471 59.75598110 - layer.0.v_cache 0.00000027 0.00068304 - layer.1.k_cache 0.00292436 6.01860310 - layer.1.v_cache 0.00000085 0.00279860 - layer.2.k_cache 0.00118067 2.07984989 - layer.2.v_cache 0.00000120 0.00431448 - layer.3.k_cache 0.00130453 2.45376319 - layer.3.v_cache 0.00000223 0.00743443 - layer.4.k_cache 0.00350570 6.33172210 - layer.4.v_cache 0.00000328 0.01212283 - layer.4.output 0.00016368 0.20549977 - ------------------------------------------------------------------------------------- - TOTAL 0.00238518 5.53494799 - (elements=5,060,608) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5060608 -Total Bytes 44268 -BPFP 0.0700 bits/point -EBPFP 0.1400 equivalent bits/point -MSE 5.534948 ----------------------- -------------------------------------------------------- -Time: 1.086s Load: 0.020s, Pack+Encode: 0.407s, Decode+Unpack: 0.659s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 353, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 353, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.5349 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample82-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample82-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample83-layer4-item1.zst (93/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample83-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 344, 128) -Output shape: (1, 344, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) -> torch.Size([1, 1, 344, 1024]) - layer.4.output: torch.Size([1, 344, 4096]) -> torch.Size([1, 1, 344, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,572B, BPFP=0.0584 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,800B, BPFP=0.0636 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,732B, BPFP=0.0848 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,272B, BPFP=0.0743 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,956B, BPFP=0.0898 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,400B, BPFP=0.0772 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,164B, BPFP=0.0946 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,608B, BPFP=0.0819 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,628B, BPFP=0.0824 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,580B, BPFP=0.0813 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,392B, BPFP=0.0590 -⌛️ [2/4] FRONTEND: Frontend time: 0.403s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.662s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 344, 128]) - layer.0.v_cache: torch.Size([1, 8, 344, 128]) - layer.1.k_cache: torch.Size([1, 8, 344, 128]) - layer.1.v_cache: torch.Size([1, 8, 344, 128]) - layer.2.k_cache: torch.Size([1, 8, 344, 128]) - layer.2.v_cache: torch.Size([1, 8, 344, 128]) - layer.3.k_cache: torch.Size([1, 8, 344, 128]) - layer.3.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.k_cache: torch.Size([1, 8, 344, 128]) - layer.4.v_cache: torch.Size([1, 8, 344, 128]) - layer.4.output: torch.Size([1, 344, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02504009 55.45518600 - layer.0.v_cache 0.00000027 0.00069183 - layer.1.k_cache 0.00288154 5.91647055 - layer.1.v_cache 0.00000082 0.00289012 - layer.2.k_cache 0.00115445 2.08764205 - layer.2.v_cache 0.00000117 0.00440848 - layer.3.k_cache 0.00131257 2.42601297 - layer.3.v_cache 0.00000224 0.00763815 - layer.4.k_cache 0.00356118 6.31783206 - layer.4.v_cache 0.00000339 0.01241674 - layer.4.output 0.00014496 0.19910544 - ------------------------------------------------------------------------------------- - TOTAL 0.00246697 5.21625791 - (elements=4,931,584) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4931584 -Total Bytes 45104 -BPFP 0.0732 bits/point -EBPFP 0.1463 equivalent bits/point -MSE 5.216258 ----------------------- -------------------------------------------------------- -Time: 1.084s Load: 0.018s, Pack+Encode: 0.403s, Decode+Unpack: 0.662s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 344, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 344, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.2163 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample83-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample83-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample85-layer4-item1.zst (94/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample85-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 321, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 321, 128) -Output shape: (1, 321, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.0.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.1.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.1.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.2.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.2.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.3.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.3.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.4.k_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.4.v_cache: torch.Size([1, 8, 321, 128]) -> torch.Size([1, 1, 321, 1024]) - layer.4.output: torch.Size([1, 321, 4096]) -> torch.Size([1, 1, 321, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,588B, BPFP=0.0630 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,672B, BPFP=0.0650 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,644B, BPFP=0.0887 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,044B, BPFP=0.0741 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,968B, BPFP=0.0966 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,240B, BPFP=0.0789 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,084B, BPFP=0.0994 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,476B, BPFP=0.0846 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,484B, BPFP=0.0848 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,416B, BPFP=0.0831 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,156B, BPFP=0.0618 -⌛️ [2/4] FRONTEND: Frontend time: 0.411s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 321, 128]) - layer.0.v_cache: torch.Size([1, 8, 321, 128]) - layer.1.k_cache: torch.Size([1, 8, 321, 128]) - layer.1.v_cache: torch.Size([1, 8, 321, 128]) - layer.2.k_cache: torch.Size([1, 8, 321, 128]) - layer.2.v_cache: torch.Size([1, 8, 321, 128]) - layer.3.k_cache: torch.Size([1, 8, 321, 128]) - layer.3.v_cache: torch.Size([1, 8, 321, 128]) - layer.4.k_cache: torch.Size([1, 8, 321, 128]) - layer.4.v_cache: torch.Size([1, 8, 321, 128]) - layer.4.output: torch.Size([1, 321, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.656s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 321, 128]) - layer.0.v_cache: torch.Size([1, 8, 321, 128]) - layer.1.k_cache: torch.Size([1, 8, 321, 128]) - layer.1.v_cache: torch.Size([1, 8, 321, 128]) - layer.2.k_cache: torch.Size([1, 8, 321, 128]) - layer.2.v_cache: torch.Size([1, 8, 321, 128]) - layer.3.k_cache: torch.Size([1, 8, 321, 128]) - layer.3.v_cache: torch.Size([1, 8, 321, 128]) - layer.4.k_cache: torch.Size([1, 8, 321, 128]) - layer.4.v_cache: torch.Size([1, 8, 321, 128]) - layer.4.output: torch.Size([1, 321, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02357271 53.11094480 - layer.0.v_cache 0.00000026 0.00066052 - layer.1.k_cache 0.00287206 5.80189738 - layer.1.v_cache 0.00000077 0.00275036 - layer.2.k_cache 0.00116438 2.11715004 - layer.2.v_cache 0.00000114 0.00434066 - layer.3.k_cache 0.00132116 2.45497621 - layer.3.v_cache 0.00000214 0.00736206 - layer.4.k_cache 0.00354096 6.38693779 - layer.4.v_cache 0.00000337 0.01238477 - layer.4.output 0.00013764 0.18975942 - ------------------------------------------------------------------------------------- - TOTAL 0.00235925 5.04703159 - (elements=4,601,856) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4601856 -Total Bytes 43772 -BPFP 0.0761 bits/point -EBPFP 0.1522 equivalent bits/point -MSE 5.047032 ----------------------- -------------------------------------------------------- -Time: 1.086s Load: 0.019s, Pack+Encode: 0.411s, Decode+Unpack: 0.656s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 321, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 321, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.0470 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample85-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample85-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample86-layer4-item1.zst (95/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample86-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,668B, BPFP=0.0634 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,792B, BPFP=0.0663 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,800B, BPFP=0.0902 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,248B, BPFP=0.0771 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,072B, BPFP=0.0967 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,484B, BPFP=0.0827 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,208B, BPFP=0.0999 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,640B, BPFP=0.0864 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,620B, BPFP=0.0860 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,580B, BPFP=0.0850 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,488B, BPFP=0.0623 -⌛️ [2/4] FRONTEND: Frontend time: 0.404s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.657s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02395787 47.98670213 - layer.0.v_cache 0.00000026 0.00066067 - layer.1.k_cache 0.00285945 5.93733786 - layer.1.v_cache 0.00000078 0.00276711 - layer.2.k_cache 0.00114467 2.08679403 - layer.2.v_cache 0.00000113 0.00434279 - layer.3.k_cache 0.00130198 2.44473684 - layer.3.v_cache 0.00000219 0.00747172 - layer.4.k_cache 0.00350851 6.30673413 - layer.4.v_cache 0.00000338 0.01243241 - layer.4.output 0.00014624 0.19502356 - ------------------------------------------------------------------------------------- - TOTAL 0.00238323 4.68357671 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 45600 -BPFP 0.0773 bits/point -EBPFP 0.1547 equivalent bits/point -MSE 4.683577 ----------------------- -------------------------------------------------------- -Time: 1.078s Load: 0.018s, Pack+Encode: 0.404s, Decode+Unpack: 0.657s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.6836 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample86-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample86-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample87-layer4-item1.zst (96/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample87-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 350, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 350, 128) -Output shape: (1, 350, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.0.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.1.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.1.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.2.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.2.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.3.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.3.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.4.k_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.4.v_cache: torch.Size([1, 8, 350, 128]) -> torch.Size([1, 1, 350, 1024]) - layer.4.output: torch.Size([1, 350, 4096]) -> torch.Size([1, 1, 350, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,460B, BPFP=0.0549 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,696B, BPFP=0.0602 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,624B, BPFP=0.0809 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,036B, BPFP=0.0678 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,892B, BPFP=0.0869 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,252B, BPFP=0.0726 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,048B, BPFP=0.0904 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,528B, BPFP=0.0788 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,504B, BPFP=0.0782 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,416B, BPFP=0.0762 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 9,972B, BPFP=0.0556 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 350, 128]) - layer.0.v_cache: torch.Size([1, 8, 350, 128]) - layer.1.k_cache: torch.Size([1, 8, 350, 128]) - layer.1.v_cache: torch.Size([1, 8, 350, 128]) - layer.2.k_cache: torch.Size([1, 8, 350, 128]) - layer.2.v_cache: torch.Size([1, 8, 350, 128]) - layer.3.k_cache: torch.Size([1, 8, 350, 128]) - layer.3.v_cache: torch.Size([1, 8, 350, 128]) - layer.4.k_cache: torch.Size([1, 8, 350, 128]) - layer.4.v_cache: torch.Size([1, 8, 350, 128]) - layer.4.output: torch.Size([1, 350, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.662s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 350, 128]) - layer.0.v_cache: torch.Size([1, 8, 350, 128]) - layer.1.k_cache: torch.Size([1, 8, 350, 128]) - layer.1.v_cache: torch.Size([1, 8, 350, 128]) - layer.2.k_cache: torch.Size([1, 8, 350, 128]) - layer.2.v_cache: torch.Size([1, 8, 350, 128]) - layer.3.k_cache: torch.Size([1, 8, 350, 128]) - layer.3.v_cache: torch.Size([1, 8, 350, 128]) - layer.4.k_cache: torch.Size([1, 8, 350, 128]) - layer.4.v_cache: torch.Size([1, 8, 350, 128]) - layer.4.output: torch.Size([1, 350, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02252116 46.64271205 - layer.0.v_cache 0.00000028 0.00066252 - layer.1.k_cache 0.00283737 5.75895020 - layer.1.v_cache 0.00000078 0.00254572 - layer.2.k_cache 0.00114986 2.04292236 - layer.2.v_cache 0.00000111 0.00400022 - layer.3.k_cache 0.00128401 2.38979231 - layer.3.v_cache 0.00000216 0.00698677 - layer.4.k_cache 0.00352053 6.28094238 - layer.4.v_cache 0.00000313 0.01160339 - layer.4.output 0.00014459 0.19411460 - ------------------------------------------------------------------------------------- - TOTAL 0.00227848 4.56554117 - (elements=5,017,600) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 5017600 -Total Bytes 43428 -BPFP 0.0692 bits/point -EBPFP 0.1385 equivalent bits/point -MSE 4.565541 ----------------------- -------------------------------------------------------- -Time: 1.086s Load: 0.018s, Pack+Encode: 0.406s, Decode+Unpack: 0.662s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 350, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 350, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.5655 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample87-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample87-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample89-layer4-item1.zst (97/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample89-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.019s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 323, 128) -Output shape: (1, 323, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) -> torch.Size([1, 1, 323, 1024]) - layer.4.output: torch.Size([1, 323, 4096]) -> torch.Size([1, 1, 323, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,588B, BPFP=0.0626 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,660B, BPFP=0.0643 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,664B, BPFP=0.0886 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,052B, BPFP=0.0738 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 3,964B, BPFP=0.0959 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,256B, BPFP=0.0788 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,088B, BPFP=0.0989 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,484B, BPFP=0.0843 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,520B, BPFP=0.0851 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,432B, BPFP=0.0830 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,172B, BPFP=0.0615 -⌛️ [2/4] FRONTEND: Frontend time: 0.404s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.656s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 323, 128]) - layer.0.v_cache: torch.Size([1, 8, 323, 128]) - layer.1.k_cache: torch.Size([1, 8, 323, 128]) - layer.1.v_cache: torch.Size([1, 8, 323, 128]) - layer.2.k_cache: torch.Size([1, 8, 323, 128]) - layer.2.v_cache: torch.Size([1, 8, 323, 128]) - layer.3.k_cache: torch.Size([1, 8, 323, 128]) - layer.3.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.k_cache: torch.Size([1, 8, 323, 128]) - layer.4.v_cache: torch.Size([1, 8, 323, 128]) - layer.4.output: torch.Size([1, 323, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02344667 64.70478788 - layer.0.v_cache 0.00000026 0.00066354 - layer.1.k_cache 0.00293942 5.83719902 - layer.1.v_cache 0.00000088 0.00278483 - layer.2.k_cache 0.00118806 2.11470244 - layer.2.v_cache 0.00000115 0.00435504 - layer.3.k_cache 0.00133183 2.46344731 - layer.3.v_cache 0.00000219 0.00753963 - layer.4.k_cache 0.00365354 6.55926117 - layer.4.v_cache 0.00000330 0.01226085 - layer.4.output 0.00014070 0.20389347 - ------------------------------------------------------------------------------------- - TOTAL 0.00236644 5.89446968 - (elements=4,630,528) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4630528 -Total Bytes 43880 -BPFP 0.0758 bits/point -EBPFP 0.1516 equivalent bits/point -MSE 5.894470 ----------------------- -------------------------------------------------------- -Time: 1.080s Load: 0.019s, Pack+Encode: 0.404s, Decode+Unpack: 0.656s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 323, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 323, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.8945 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample89-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample89-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample9-layer4-item1.zst (98/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample9-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 419, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.022s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 419, 128) -Output shape: (1, 419, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.0.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.1.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.1.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.2.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.2.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.3.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.3.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.4.k_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.4.v_cache: torch.Size([1, 8, 419, 128]) -> torch.Size([1, 1, 419, 1024]) - layer.4.output: torch.Size([1, 419, 4096]) -> torch.Size([1, 1, 419, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,892B, BPFP=0.0539 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 3,092B, BPFP=0.0577 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 4,228B, BPFP=0.0788 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,512B, BPFP=0.0655 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,528B, BPFP=0.0844 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,760B, BPFP=0.0701 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,744B, BPFP=0.0885 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 4,056B, BPFP=0.0756 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 4,104B, BPFP=0.0765 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,972B, BPFP=0.0741 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 11,172B, BPFP=0.0521 -⌛️ [2/4] FRONTEND: Frontend time: 0.464s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 419, 128]) - layer.0.v_cache: torch.Size([1, 8, 419, 128]) - layer.1.k_cache: torch.Size([1, 8, 419, 128]) - layer.1.v_cache: torch.Size([1, 8, 419, 128]) - layer.2.k_cache: torch.Size([1, 8, 419, 128]) - layer.2.v_cache: torch.Size([1, 8, 419, 128]) - layer.3.k_cache: torch.Size([1, 8, 419, 128]) - layer.3.v_cache: torch.Size([1, 8, 419, 128]) - layer.4.k_cache: torch.Size([1, 8, 419, 128]) - layer.4.v_cache: torch.Size([1, 8, 419, 128]) - layer.4.output: torch.Size([1, 419, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.760s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 419, 128]) - layer.0.v_cache: torch.Size([1, 8, 419, 128]) - layer.1.k_cache: torch.Size([1, 8, 419, 128]) - layer.1.v_cache: torch.Size([1, 8, 419, 128]) - layer.2.k_cache: torch.Size([1, 8, 419, 128]) - layer.2.v_cache: torch.Size([1, 8, 419, 128]) - layer.3.k_cache: torch.Size([1, 8, 419, 128]) - layer.3.v_cache: torch.Size([1, 8, 419, 128]) - layer.4.k_cache: torch.Size([1, 8, 419, 128]) - layer.4.v_cache: torch.Size([1, 8, 419, 128]) - layer.4.output: torch.Size([1, 419, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02248870 53.86108107 - layer.0.v_cache 0.00000028 0.00065916 - layer.1.k_cache 0.00282601 5.78421989 - layer.1.v_cache 0.00000075 0.00247741 - layer.2.k_cache 0.00117249 2.03530629 - layer.2.v_cache 0.00000110 0.00385421 - layer.3.k_cache 0.00129145 2.39286713 - layer.3.v_cache 0.00000206 0.00681536 - layer.4.k_cache 0.00360397 6.55360373 - layer.4.v_cache 0.00000307 0.01139602 - layer.4.output 0.00013808 0.18877058 - ------------------------------------------------------------------------------------- - TOTAL 0.00228159 5.10052590 - (elements=6,006,784) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 6006784 -Total Bytes 50060 -BPFP 0.0667 bits/point -EBPFP 0.1333 equivalent bits/point -MSE 5.100526 ----------------------- -------------------------------------------------------- -Time: 1.246s Load: 0.022s, Pack+Encode: 0.464s, Decode+Unpack: 0.760s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 419, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 419, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 5.1005 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample9-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample9-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample93-layer4-item1.zst (99/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample93-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 336, 128) -Output shape: (1, 336, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) -> torch.Size([1, 1, 336, 1024]) - layer.4.output: torch.Size([1, 336, 4096]) -> torch.Size([1, 1, 336, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,572B, BPFP=0.0598 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,840B, BPFP=0.0660 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,808B, BPFP=0.0885 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,312B, BPFP=0.0770 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,000B, BPFP=0.0930 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,508B, BPFP=0.0816 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,208B, BPFP=0.0978 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,712B, BPFP=0.0863 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,632B, BPFP=0.0844 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,648B, BPFP=0.0848 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,288B, BPFP=0.0598 -⌛️ [2/4] FRONTEND: Frontend time: 0.412s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.666s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 336, 128]) - layer.0.v_cache: torch.Size([1, 8, 336, 128]) - layer.1.k_cache: torch.Size([1, 8, 336, 128]) - layer.1.v_cache: torch.Size([1, 8, 336, 128]) - layer.2.k_cache: torch.Size([1, 8, 336, 128]) - layer.2.v_cache: torch.Size([1, 8, 336, 128]) - layer.3.k_cache: torch.Size([1, 8, 336, 128]) - layer.3.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.k_cache: torch.Size([1, 8, 336, 128]) - layer.4.v_cache: torch.Size([1, 8, 336, 128]) - layer.4.output: torch.Size([1, 336, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02369028 47.16384452 - layer.0.v_cache 0.00000026 0.00066473 - layer.1.k_cache 0.00291391 6.04119146 - layer.1.v_cache 0.00000078 0.00278952 - layer.2.k_cache 0.00115383 2.09402557 - layer.2.v_cache 0.00000117 0.00434630 - layer.3.k_cache 0.00133386 2.44434266 - layer.3.v_cache 0.00000214 0.00735969 - layer.4.k_cache 0.00353874 6.37219311 - layer.4.v_cache 0.00000342 0.01234078 - layer.4.output 0.00014590 0.19556191 - ------------------------------------------------------------------------------------- - TOTAL 0.00237300 4.63752471 - (elements=4,816,896) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4816896 -Total Bytes 45528 -BPFP 0.0756 bits/point -EBPFP 0.1512 equivalent bits/point -MSE 4.637525 ----------------------- -------------------------------------------------------- -Time: 1.097s Load: 0.018s, Pack+Encode: 0.412s, Decode+Unpack: 0.666s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 336, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 336, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.6375 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample93-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample93-layer4-item1.zst - - 💪 Processing: ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample97-layer4-item1.zst (100/100) - -[1/4] FRONTEND: Loading features from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample97-layer4-item1.zst... - -Original data structure: -root: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -⌛️ [1/4] FRONTEND: Load time: 0.018s - ------------------------------------------------------------- -Qwen Features Summary ------------------------------------------------------------- -Number of layers: 5 -Layer indices: [0, 1, 2, 3, 4] -Last layer index: 4 -Cache shape: (1, 8, 329, 128) -Output shape: (1, 329, 4096) -Data type: torch.bfloat16 ------------------------------------------------------------- - -[2/4] FRONTEND: Pack + Encode (strategy: individual)... - IndividualPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) -> torch.Size([1, 1, 329, 1024]) - layer.4.output: torch.Size([1, 329, 4096]) -> torch.Size([1, 1, 329, 4096]) - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - layer.0.k_cache: 2,676B, BPFP=0.0635 - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - layer.0.v_cache: 2,772B, BPFP=0.0658 - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - layer.1.k_cache: 3,792B, BPFP=0.0900 - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - layer.1.v_cache: 3,236B, BPFP=0.0768 - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - layer.2.k_cache: 4,064B, BPFP=0.0965 - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - layer.2.v_cache: 3,404B, BPFP=0.0808 - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - layer.3.k_cache: 4,224B, BPFP=0.1003 - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - layer.3.v_cache: 3,684B, BPFP=0.0875 - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - layer.4.k_cache: 3,648B, BPFP=0.0866 - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - layer.4.v_cache: 3,588B, BPFP=0.0852 - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - layer.4.output: 10,492B, BPFP=0.0623 -⌛️ [2/4] FRONTEND: Frontend time: 0.406s (Pack+Encode) - -[3/4] BACKEND: Decode + Unpack... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) -⌛️ [3/4] BACKEND: Backend time: 0.662s - -[4/4] METRICS: Computing MSE Breakdown... - Using per-key quantization points (layer.0.k_cache: torch.Size([256])) for layer.0.k_cache - Using per-key quantization points (layer.0.v_cache: torch.Size([256])) for layer.0.v_cache - Using per-key quantization points (layer.1.k_cache: torch.Size([256])) for layer.1.k_cache - Using per-key quantization points (layer.1.v_cache: torch.Size([256])) for layer.1.v_cache - Using per-key quantization points (layer.2.k_cache: torch.Size([256])) for layer.2.k_cache - Using per-key quantization points (layer.2.v_cache: torch.Size([256])) for layer.2.v_cache - Using per-key quantization points (layer.3.k_cache: torch.Size([256])) for layer.3.k_cache - Using per-key quantization points (layer.3.v_cache: torch.Size([256])) for layer.3.v_cache - Using per-key quantization points (layer.4.k_cache: torch.Size([256])) for layer.4.k_cache - Using per-key quantization points (layer.4.v_cache: torch.Size([256])) for layer.4.v_cache - Using per-key quantization points (layer.4.output: torch.Size([256])) for layer.4.output - IndividualUnPacker: - layer.0.k_cache: torch.Size([1, 8, 329, 128]) - layer.0.v_cache: torch.Size([1, 8, 329, 128]) - layer.1.k_cache: torch.Size([1, 8, 329, 128]) - layer.1.v_cache: torch.Size([1, 8, 329, 128]) - layer.2.k_cache: torch.Size([1, 8, 329, 128]) - layer.2.v_cache: torch.Size([1, 8, 329, 128]) - layer.3.k_cache: torch.Size([1, 8, 329, 128]) - layer.3.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.k_cache: torch.Size([1, 8, 329, 128]) - layer.4.v_cache: torch.Size([1, 8, 329, 128]) - layer.4.output: torch.Size([1, 329, 4096]) - Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): - Key Quant-MSE Total-MSE - ------------------------------------------------------------------------------------- - layer.0.k_cache 0.02541956 47.46758644 - layer.0.v_cache 0.00000027 0.00067297 - layer.1.k_cache 0.00292153 5.94476671 - layer.1.v_cache 0.00000078 0.00284235 - layer.2.k_cache 0.00117184 2.13225502 - layer.2.v_cache 0.00000114 0.00438225 - layer.3.k_cache 0.00131754 2.44949832 - layer.3.v_cache 0.00000217 0.00751555 - layer.4.k_cache 0.00353394 6.42612483 - layer.4.v_cache 0.00000342 0.01239105 - layer.4.output 0.00014204 0.18867147 - ------------------------------------------------------------------------------------- - TOTAL 0.00249574 4.65733724 - (elements=4,716,544) ----------------------- -------------------------------------------------------- -SAMPLE-WISE STATISTICS ----------------------- -------------------------------------------------------- -Handler qwen -Strategy individual -Architecture hyperprior-featurecoding ----------------------- -------------------------------------------------------- -Total Elements 4716544 -Total Bytes 45580 -BPFP 0.0773 bits/point -EBPFP 0.1546 equivalent bits/point -MSE 4.657337 ----------------------- -------------------------------------------------------- -Time: 1.086s Load: 0.018s, Pack+Encode: 0.406s, Decode+Unpack: 0.662s ----------------------- -------------------------------------------------------- -Restored Feature Format: [dict] with 3 keys - key['output']: [Tensor] shape=torch.Size([1, 329, 4096]), dtype=torch.bfloat16, device=cpu - key['key']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - key['value']: [list] with 5 items - item[0]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[1]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[2]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[3]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu - item[4]: [Tensor] shape=torch.Size([1, 8, 329, 128]), dtype=torch.bfloat16, device=cpu -💾 Converting with 4.6573 MSE: - from ../datasets/Qwen3-500features-L5wCache/Qwen3-500features-L5wCache/qwen/qwen3-8b/fc_hellaswag/sample97-layer4-item1.zst - to output-fixed/qwen/lambda0.001/hyperprior-featurecoding-8bit-individual/fc_hellaswag/sample97-layer4-item1.zst ------------------------- ---------------------------- -TOTAL PROCESSING SUMMARY ------------------------- ---------------------------- -Total files 100 -Avg BPFP 0.0721 bits/point -Avg EBPFP 0.1443 equivalent bits/point -Avg MSE 5.090933 -Avg Time 1.075s ------------------------- ---------------------------- +version https://git-lfs.github.com/spec/v1 +oid sha256:da0a2c85466f8508412fed703c16fbf6b0f4a09654ff6f5a04900d1fe06893c4 +size 1123813