diff --git a/checksums/extended-evidence-v1.sha256 b/checksums/extended-evidence-v1.sha256 new file mode 100644 index 0000000000000000000000000000000000000000..67dc11f2385235c409aee591e628826e7c3a0cc9 --- /dev/null +++ b/checksums/extended-evidence-v1.sha256 @@ -0,0 +1,309 @@ +2469e607e691cba884beda0fe55fe69bf1dbb14ca8f22765a81139f6bf60d18d docs/extended-evidence-v1/EXCLUSIONS.md +41d8ad21168e4958be51f2a6649c585c2151c807d95f6d05daaef0d837ee0404 docs/extended-evidence-v1/README.md +70082850ca060f1a2b0a7762c85ba07d515b4c66d87382870dda0be5a5ceafd1 figures/data/qwen3-4b/v1/q3_4b_armA_traj.csv +6f406c21eea7ee25bda9c00bec301d705a3427bf2de2e6ab839ef31cc701090a figures/data/qwen3-4b/v1/q3_4b_armB_traj.csv +5dd37b6c7d9a3d4c6ede536aec2a869aa26afd8c67c3096a04e66355827815e9 figures/data/qwen3-4b/v1/q3_4b_gen_traj.csv +89992f95494655a1263b1f17c098c97e68bba6aa19279f29b7d22a049096ce50 figures/rendered/historical/v1/code250k_vs_gen.png +66e0cfc000e9583110ce5d174ab1874210eee0b026d25fd863e7f69139f78ec3 figures/rendered/historical/v1/fig_bigdata_vs_gen.png +c88b5a91d84cfbf30f5f9b1813ae470f808e348d27f224f57b8704490f839016 figures/rendered/historical/v1/fig_exp1_forgetting.png +2b9633ec3524f23766a72e788206c114e02d7b0da5be8f3cf8a501848fe2fb52 figures/rendered/historical/v1/fig_exp1_three_recipes.png +e8602bc3791560060deba3ad715eedf6fd9d1ef5d76c1774d3ef5a5fb0ccb2bd figures/rendered/historical/v1/fig_exp4_inference.png +bf243131b8c61f12c258760307178a8d7888e41c5ef1f5a8981f37b7f639b83a figures/rendered/historical/v1/fig_exp5_method_vs_data.png +58be85cb2527aa44c3e8c2579831138104bc70fa7e33794e1855df76988e9d6d figures/rendered/historical/v1/fig_forgetting_matrix.png +73c7ceff7bf8f01c0bc3e861c689532619dc2f358f51b37a0d9cedd8b07779bc figures/rendered/historical/v1/fig_fusion_combined.png +cc68719d712a0bc37321c46ec7d9b6368c08fc3f4523ef1d1c3d7771e5a99a37 figures/rendered/historical/v1/fig_gen800k_loss_al.png +ca381bfb16abfde13f5e88c917b27698155fa073e99cba70ad50235c6565a155 figures/rendered/historical/v1/fig_merged_vs_specialist.png +16825387c4de0251447e50ba54aa99585cf8fc2790b1d0f4c8438035c40e23f0 figures/rendered/historical/v1/fig_q3_4b_traj.png +77ae4d2ff2f302212f1252b8bd1fbdc3c0d244dfd2a1be06d093a964575eff8e figures/rendered/historical/v1/fig_serving.png +1b8928ad66e0c86b58bc8d9088af11647e78f4f42180e95a0a5bd96017f623e3 figures/rendered/historical/v1/fig_smalldata_vs_gen.png +0e3e6ae946db8f2922832f3b0b09673877c0cf68a55f9027fdc93d60399b9446 figures/rendered/historical/v1/fig_specialist_overall.png +861c7f202252b6798391fb5bc98bc88d7379e38b3c7d77d2e4fbf0f74f144085 figures/rendered/historical/v1/fig_warm_saturation.png +eb2084ef355799b98ab08000994e54aa8966196cbb3485235a942ed8966ac32c figures/rendered/historical/v1/gen_per_domain_AL.png +2f484b361bed4165c4aac24a52ad66b7a3f9ea7948d7f617c764784085ac39b5 figures/rendered/historical/v1/gen_vs_merge_attn.png +221d21d5085f280ace83fe2532a3ae52858a6787af5f28addfd480095114984b figures/rendered/historical/v1/joint_vs_pipeline_code.png +7bb2f21888c8c79e731add5c61021aff4be2c58a9dfc499601d902700867802c figures/rendered/historical/v1/overfit_attention_ladder.png +d8b24897fdce78a542aa0f9387e13299ffe3a34c679b607470ac984bddbbf10c figures/rendered/historical/v1/overfit_loss_al.png +77b4ad8a26a01b67b052d7374de6b11350d89ba6eace1e8d135af159ffca5c7f figures/rendered/historical/v1/reasonmix_5arm_per_domain.png +7c314a567c8cadfcfb602bc0ad10c328b6dca571213b88ef656bf78404350af4 figures/rendered/historical/v1/recipe_budget.png +0415d8aa007c14b576bb25e56837b95235cf9e68afe828a775be58f265b46cb9 figures/rendered/paper-current-candidate/v1/fig_architecture_v8.drawio +3961c8a6ccc80992197b8746453cffd1d6f87ed158ce6103a72b4db361996f66 figures/rendered/paper-current-candidate/v1/fig_architecture_v8.png +0a8c69fabf7ca2a57a13b4dd77edf2b94bfa1fd31fd9f28a259f7b014ec07048 figures/rendered/paper-current-candidate/v1/fig_architecture_v8.svg +dc224065a687b7631cec8da84803ec6b8b1c5c34b737dfaa4e7077c9eb04318d figures/rendered/paper-current-candidate/v1/fig_main_results.png +c843a6bd2b0a20aad7d6b2816277b60af03b44bfd4a1a8a75bc7ac1c74bac746 figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.pdf +1243e3db541c0be5dbdacdabf6046ad065911ec2080bacd7d6e69bc9100689db figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.png +da25818d68a3da676fed16a019beadd6b12f5e4d59ade85840c23a6070240323 licenses/extended-evidence-v1/LICENSE +01369f893c8ca40e0b1bfefc3d0a006833bd1757dc287755d1dcd9bd6e5fd7bc manifests/extended-evidence-v1/artifact-manifest.jsonl +ea5920812d91ae4a6e6cc503936d1c87fb833b622ef2b5bb9fb24037a4e5195e manifests/extended-evidence-v1/release-summary.json +fe3d7a8074dc76533eefa49e627d779650c811fb0068698546330d811b8cb3c5 recipes/plotting/v1/build_mos_architecture_drawio.py +c8e6297def7e8dfc361a437c798c8b8586f6f0ee99772cfa915ecc325ea9ba01 recipes/plotting/v1/fig_5arm_per_domain.py +c6ec2787808a900478e9a92fb7f3092e04835603882d39a3ad0d24b2d486e193 recipes/plotting/v1/fig_code250k_vs_gen.py +f49d9a5dfc3f7512af3cc86f0729774344357fa7de748e9917e90a5a4bf16d84 recipes/plotting/v1/fig_exp1_forgetting.py +9fd98d1f1bb1a6e449bf503569fd8fbd82d89f7366a811f0fd9be4e3b9ac7e95 recipes/plotting/v1/fig_exp1_three_recipes.py +ebe9bc1be8f0de7bdd83bcc74e71314fd9ca7758802bfa660128a265c5e4f9f1 recipes/plotting/v1/fig_exp4_inference.py +44f959ce1232a19ea3dcee93bb41df5268171e47cf2d66ed4b5e7d84312497a5 recipes/plotting/v1/fig_exp5_method_vs_data.py +39542922ee86c359fc3c6e0153daba478ada98e8d10da5e40b8b9994a440a87f recipes/plotting/v1/fig_fusion_combined.py +6967a36b0a4ce87c5650409a561bc90a186eb25633501593c87e6da9460c2bd2 recipes/plotting/v1/fig_matrix_and_arms.py +6eb66c23518bd7e2b8edc74f23ddb53635cac18ae839c4862ec9bfbc1ca8bdc5 recipes/plotting/v1/fig_merged_vs_specialist.py +a692b7e495c6d3ceaf46857436a115ffc299375936e37f820ddda02468a724b7 recipes/plotting/v1/fig_serving_specialist.py +d28b0d29bda5f5ee10c4ea22fa24e7b1b07e5ba2d86c57f74f58285b06e9b58e recipes/plotting/v1/plot_main_results.py +663f813697a0026b093a0a6ef68d3b99be3a7c159eadbe2c5b3522e645d92eec recipes/plotting/v1/plot_mos_5x5_gains.py +6fdf9f53c384d1ec47ed182dcf2d44987ff30bb0255bef0061fa7c723b785c0e recipes/plotting/v1/plot_qwen3_4b_matched.py +a7beb62c3b84008553ff9051ecc76bbd859667bf28898b182317363c8128e404 recipes/plotting/v1/plot_recipe_budget.py +b7ebb2c81dacb6ef06dcb87fb879ea663f2327b7417606520bb30cb299ef006a recipes/requirements-v1.txt +dae17f81b9fa4b5dba14545e732f6731e359e80e3845f937c09fe81ef8a6e068 results/historical-al-curves/v1/a2_code/al_curve.csv +7d56cbca48fdec9cc37fc26e701ca5da6841884f8a49e51c929090c49cb5fe6b results/historical-al-curves/v1/a2_creative_writing/al_curve.csv +5da7a3f0e5d955881ce5903f4a7b7567212f89400f130a574fdc0cb783169551 results/historical-al-curves/v1/a2_factual_qa/al_curve.csv +aee9cf5c10f54cadba4b30b513c18b2651f1fbc6b10d25da719eaea74f471682 results/historical-al-curves/v1/a2_general/al_curve.csv +4ce95c7937d1362f0593ad93531d65620a484e138237ce793ea1b3ef03099c5b results/historical-al-curves/v1/a2_math/al_curve.csv +e88672fadfaaaeb1088504c919208f8b125f1aecb7a14942fc5888d572714857 results/historical-al-curves/v1/a3_code/al_curve.csv +b5a492456138dc23d09264c26ca671082eb3fb36765ec5398aa31f56d77d2cc1 results/historical-al-curves/v1/a3_creative_writing/al_curve.csv +b7b8776a83582db069ecd349784aa52d40dcd3bbd6c100c13780a873590c4b98 results/historical-al-curves/v1/a3_factual_qa/al_curve.csv +4073a14f31f2d26f0c1974f3e571a45c24209bf2562b2c7725d207d134047a9e results/historical-al-curves/v1/a3_general/al_curve.csv +07ba1e5ce58ff013aab9a96738301295f32731facc3ec5b7678802d5ef8c996f results/historical-al-curves/v1/a3_math/al_curve.csv +cd427bc83af9d2f892924bd1993ead327acab61c06279e5c627abe2e18cc6809 results/historical-al-curves/v1/a4_code/al_curve.csv +c1867fc99b121cfd613ce2a9d40f8f12a13491e146f85d3175d7bf8c3591c02f results/historical-al-curves/v1/a4_creative_writing/al_curve.csv +cf142156c17e74d5ac67967b93995dbaf47ee34d4ab28cf65d0044846cd52785 results/historical-al-curves/v1/a4_factual_qa/al_curve.csv +81d3dcaacdd7405966d5eb5715dd04c434e714abf3a9620ae492430065c8dbc6 results/historical-al-curves/v1/a4_general/al_curve.csv +8e717efab00e869e764262597696297a9bfd5cd7839e0d9ddd2ba623acb0aa05 results/historical-al-curves/v1/a4_math/al_curve.csv +c105111e04a724b5b21af90fd5e1ecd2b10820db32c9fb041e3cfe153e862f6f results/historical-al-curves/v1/a5_code/al_curve.csv +454d8a6f5a03494e4d8aace991965482a858e04bbcdd4c16cb07773ba3895afa results/historical-al-curves/v1/a5_creative_writing/al_curve.csv +43df7e88a7ddd7bbbc9471c65682b4584af117b0d1f02dc0ade07bb29395c9cf results/historical-al-curves/v1/a5_factual_qa/al_curve.csv +e64d977591372dca44bb442ced4a6f5c1f39464a84471bf1ebc21abe5e29bdd3 results/historical-al-curves/v1/a5_general/al_curve.csv +b85e52e687ba97220b8cad613cc209578df53767e40ab55fd9cf2fdf5808f3b1 results/historical-al-curves/v1/a5_math/al_curve.csv +0784056349eac36c44dc1da700b5e051dd60ebfb832254fadf0e30049bdb6644 results/historical-al-curves/v1/a6_code/al_curve.csv +6f0c6689cc637a4046f1cd36d32c97f5df4d6931d4bbd0df7f373929a7384258 results/historical-al-curves/v1/a6_creative_writing/al_curve.csv +8ab6645dd36b4c7ed6e150591aa796d07fca2f453083a4ea7690b19d14f0522c results/historical-al-curves/v1/a6_factual_qa/al_curve.csv +dcc0ad60fb88da0a1c2b124b7e6daa94fa4bea76de8ac393eac3edefd900de8d results/historical-al-curves/v1/a6_general/al_curve.csv +743758c18a447dc760c512d1525a6adea57a0f8086d0d0b7b8acf0a88a9b16b7 results/historical-al-curves/v1/a6_math/al_curve.csv +c3c3784e3a7046135fe0772b9e9acedbfe45473bf125a7c69766e08118a90b9f results/historical-al-curves/v1/a7_code/al_curve.csv +0eab7d6c41d11ccd0653115a7077d4b9e22943bbd9512867b71eaa2e0df01327 results/historical-al-curves/v1/a7_creative_writing/al_curve.csv +d86f5a9a467df89b88deb48595e53915114626e692200afe2b1cceb68951f7ad results/historical-al-curves/v1/a7_factual_qa/al_curve.csv +c66d5d71f44d502be2f4319b57d99d7ab498e470e8b29808fc7943c5cde07423 results/historical-al-curves/v1/a7_general/al_curve.csv +07fa6f01a556aff84978b938d7210e2c32cbcea0ff28b672895169ef2096b9fa results/historical-al-curves/v1/a7_math/al_curve.csv +63bc1738955954356859cdca199f48a3948003f6c6ceb523e7658dc6297c4fe5 results/historical-al-curves/v1/aw0_code/al_curve.csv +58a7c10f203fdfca83c4445aab6b8db5bc93a5ebf6c5ee39a30adea955238048 results/historical-al-curves/v1/aw1_code/al_curve.csv +766d1a00823d6db3f2cd6a3d96e11b935d31120f95f780b6d1f8b41a53a95f82 results/historical-al-curves/v1/aw2_code/al_curve.csv +b3b6db437eb990291582e1fe8180fb77a5fc44394ae3a4bd595b58a1fa95116e results/historical-al-curves/v1/aw3_code/al_curve.csv +71a499635ea64fb0e45ea136aafe69f2466b508e0b66e6b8158a681ab0fb6ffb results/historical-al-curves/v1/aw4_code/al_curve.csv +e135392f04e9db93600e871c8a82b556b7f380c7622db82489bf158358fb0a29 results/historical-al-curves/v1/aw5_code/al_curve.csv +23bdd93d5bfa5c24aed9089fb1f1903927a2de4bf26629ee42194d31994fe2c3 results/historical-al-curves/v1/aw6_code/al_curve.csv +793994b7cbf0695dcc184bf095697988aef78be0ad6c8fabb615c2f16200d652 results/historical-al-curves/v1/aw7_code/al_curve.csv +ed2bab825dda5d70ddb6a3e16a0a86633178f1c8e3a2149031d56f11be4b05b0 results/historical-al-curves/v1/b0_code/al_curve.csv +f56a485034b29f7e1c65c8021efe18f5980d732c4e1556d6550e93a52d55f616 results/historical-al-curves/v1/b0_creative_writing/al_curve.csv +f33c9d7fa3985e39ac50ff1d8dad02d72e7276a08c8d43492627be23ed67c579 results/historical-al-curves/v1/b0_factual_qa/al_curve.csv +51ea0ab13244e62257a02239ca42aade659a535afff79faafc4f37c4d6ca5316 results/historical-al-curves/v1/b0_general/al_curve.csv +405dd523d42728d330a5a5de80f12957a15f954c41f06474c0f898f841c7b4b9 results/historical-al-curves/v1/b0_math/al_curve.csv +97932cc47a49befbf27f7957d827a37c3b93d691d8b5692562c9fe52d964d941 results/historical-al-curves/v1/b1_code/al_curve.csv +f8185aae5e4724c5a86c6adfeb34c16e7767f929d3d633c0d60cab0f8127128b results/historical-al-curves/v1/b1_creative_writing/al_curve.csv +c2031c105435638bb133232dc1bd0c8b23454a368fa6bddc457426981dfbde56 results/historical-al-curves/v1/b1_factual_qa/al_curve.csv +0967339a7d6711ef84490506221730589487c1d21586cd1e131bb4b5e0540a11 results/historical-al-curves/v1/b1_general/al_curve.csv +1fb5f0f30ded6faad7a1f8a02050853cb500df707853dfbfe4239016e625a6bf results/historical-al-curves/v1/b1_math/al_curve.csv +784723990cdf7fb9cbae73598ade59da5c5e494666187d39ccab98e2a13e843c results/historical-al-curves/v1/b2_code/al_curve.csv +2d7c6e582ced15922331dbd31ca93ac17476daa07af395bba9f2c4e90793ada8 results/historical-al-curves/v1/b2_creative_writing/al_curve.csv +8a7fc49463e125c3572f43300ae6f51229d3b7a9460f8a4cd7415f4029155e0d results/historical-al-curves/v1/b2_factual_qa/al_curve.csv +17a09873412bd568cc0e707da3502f5413a58fc10085a640d4e6433645ad0307 results/historical-al-curves/v1/b2_general/al_curve.csv +8cfcdd3707b7474195e9c1d55361ebf1d7798eccc1bd5d2e77a5e354a05e0272 results/historical-al-curves/v1/b2_math/al_curve.csv +e4f9520174ba9ba33873c3c667883034cb9112e0ed8fb1e9ec6f680142ded0c5 results/historical-al-curves/v1/b2_v3_cont_curve_8x5x1x6/al_curve.csv +c36f8911c0042f6f686cd63a43330bb2dcb007709aa8b15807eb8a32dd259341 results/historical-al-curves/v1/b2_v3_curve_8x5x1x6/al_curve.csv +0944a8604c0bd5a5218a9a0c6a4b3ed370657b300e8941d9a5265f5b21634b54 results/historical-al-curves/v1/b3_code/al_curve.csv +7f35dd8bac7bec1c7fd38eee49ef144cf8cd1ea6577c28ebf7b02008a0e018c4 results/historical-al-curves/v1/b3_creative_writing/al_curve.csv +393b855f4794d72ab7fd16d2dff264c975f3407f3b781831465fd79334e705f3 results/historical-al-curves/v1/b3_factual_qa/al_curve.csv +76db113c70a264f7f3d96bb764b2befdcbd8ddd30593babbd13be47c7ccd36c2 results/historical-al-curves/v1/b3_general/al_curve.csv +cb2c4c321389e1f978648ac9b634532606885924336ee9eff3e9ee3df0dd8806 results/historical-al-curves/v1/b3_math/al_curve.csv +20dcba4e0f4a97d64d795ab5e97e377d14f083ce3ca8e9febf6ab1def6bd05d8 results/historical-al-curves/v1/b4_code/al_curve.csv +cfb4e10bdb3f8896f321403b653c01f8edc7be518376a7508ee1fe16abf5133b results/historical-al-curves/v1/b4_creative_writing/al_curve.csv +6889a3c240b28f1708a8aa1844fbd9e016ede9cd06201d3f3444e852d6bd18a9 results/historical-al-curves/v1/b4_factual_qa/al_curve.csv +93e2192320c39883bb6e536275fbc5af9992c285856a98c51f745a6c2cd322f1 results/historical-al-curves/v1/b4_general/al_curve.csv +897845f483d8fbbc7f9cf587698bc4b7a96e126f7716172bc5547f7138ba7bfd results/historical-al-curves/v1/b4_math/al_curve.csv +092f185463e80a3f3870785173b7047d62662919946293db45b2e432318717e5 results/historical-al-curves/v1/b5_code/al_curve.csv +0c193663d6bf65c6ffede829949cb80bc68801c4b52e7b3b254c410cb476e8d3 results/historical-al-curves/v1/b5_creative_writing/al_curve.csv +bd5d632a8da9392f4ff22a73c1540e8990ba72407c09eb80e5f326abd31657d3 results/historical-al-curves/v1/b5_factual_qa/al_curve.csv +7ab5552d57bee9349fa05b10f0ec20751db6d57745f192b0fff040df83c70db7 results/historical-al-curves/v1/b5_general/al_curve.csv +601ddb84962c2e9a12c49338d3688b534b1b6e8efc3ea1f7c47dfd8274ac3e85 results/historical-al-curves/v1/b5_math/al_curve.csv +d5858dc1897c4d64feadf087ea74740c3657b736e50b1561e442c7d7e66a0b3a results/historical-al-curves/v1/b6_code/al_curve.csv +752306586324989229b99b0eb4dc42d52faca211e44337bd279f063ee27c31c8 results/historical-al-curves/v1/b6_creative_writing/al_curve.csv +97e815c6b08c33cd7c10581fed459dfe06d2a01328d1f433908f4ec0261b469f results/historical-al-curves/v1/b6_factual_qa/al_curve.csv +9da6e16cbbe001a4826ad54233041c396ce4dce90e4074639f6a74ce9c4dcb62 results/historical-al-curves/v1/b6_general/al_curve.csv +66c4a1fc30f00ec6a172c79df7bc9bfd5eb20300b4b45961387b215a3459f223 results/historical-al-curves/v1/b6_math/al_curve.csv +9caaf63ef828990870bcfbc53c48d849cfd1b9a7db2f33b028a82f32e7246ccd results/historical-al-curves/v1/b7_code/al_curve.csv +16228b3deb7f7e45f190ef229186df928daaa0a182b9a7f7b5a8fe0e0bf84748 results/historical-al-curves/v1/b7_creative_writing/al_curve.csv +6d380638c474c93a128897ea0d5e38507c8fc639cd5afdac85cfbf55e9be72ae results/historical-al-curves/v1/b7_factual_qa/al_curve.csv +4c00e7142537032150da6be745d880fb00fc63e7722d3f016cc7801bc9180884 results/historical-al-curves/v1/b7_general/al_curve.csv +ea0b445042ac886293657641a64bb039fd6e2cd26a6c2b73bdf3b6446b05e567 results/historical-al-curves/v1/b7_math/al_curve.csv +4c7950c6ca3dbe06d5674d2568b5b183b56de92bf4fa76230af5c5d045e1eabd results/historical-al-curves/v1/c0_code/al_curve.csv +b74b5f158b2f61a23d899aec307d8615f8a33d158032b2a896804b0bf1450514 results/historical-al-curves/v1/c0_creative_writing/al_curve.csv +b4e8a4ec59a51f14f79275cb1341280656a6bee50653fe497c88680af5acbcee results/historical-al-curves/v1/c0_factual_qa/al_curve.csv +0ee93e536ed44dbcfab2d08f8f8a76e6590b4f35cc8fc1a88cb58248700b2a78 results/historical-al-curves/v1/c0_general/al_curve.csv +607c0206b4b6e6985297e22ee36d60fc8d315d5275b5b1d10546d2ec5de1b44e results/historical-al-curves/v1/c0_math/al_curve.csv +d7b3403ff3fe7e8cc6e119f046911c339bc297fb9d6484ea667326c5b4993f52 results/historical-al-curves/v1/c1_code/al_curve.csv +954ff064cf4cec980f9439cd5c7434272ffc93ecd38f0eb7fba1c7c08c1fa518 results/historical-al-curves/v1/c1_creative_writing/al_curve.csv +db592b66feeeb1b807c0cd192b474eb6beca4b3be00d1cf9bd1faafd6175b47a results/historical-al-curves/v1/c1_factual_qa/al_curve.csv +803576dbe5083e7b3a487a9780cb06bb58c63e82601156afe931fd3e04d7155d results/historical-al-curves/v1/c1_general/al_curve.csv +eca7f51ff784366375d5e29799fc5bf316a9ef803812d14e22c066f508c1cb35 results/historical-al-curves/v1/c1_math/al_curve.csv +65685989c83c8cde6179b3baf8725a231c5d020e326accc2f4ccb65db4ebecb5 results/historical-al-curves/v1/c2_code/al_curve.csv +868681f811ac8721811740d3c504f35bf37d1560d67d045c86e68023b7ea3dcd results/historical-al-curves/v1/c2_creative_writing/al_curve.csv +d5b2c7f25d9bcfb1ee072c5529e8cdfa8122de16cf8ae5b25601a27ea4b6b684 results/historical-al-curves/v1/c2_factual_qa/al_curve.csv +211aca725b163fc8c6db2956352f254e03005204551d0bce4fe2c11fc24166ca results/historical-al-curves/v1/c2_general/al_curve.csv +ba923a05b754e4009f41f452da060bf128b4af1d97b5d60c69cb0adf970964b4 results/historical-al-curves/v1/c2_math/al_curve.csv +7a7032f0fdb54aff1bb68456bcc4619107572d9e1c7be9f04653dffe903e2c31 results/historical-al-curves/v1/c3_code/al_curve.csv +dd88bd7aa0c7d70467c88eba8ea31c26f4db5aa82c6c485f723f1bc6a8a47f7a results/historical-al-curves/v1/c3_creative_writing/al_curve.csv +8790230d20b4996547849d1fee887ccf70a036fb69c3377c93ef237b74e9576f results/historical-al-curves/v1/c3_factual_qa/al_curve.csv +d1586a3cf87eee78a83d6fdac95a70203c4b18fa56f371b1f3f1b801fe1c27b1 results/historical-al-curves/v1/c3_general/al_curve.csv +4ad7ce036023a5644f713848b56b428fd78ba427e77ab2aea8039105c12cc5de results/historical-al-curves/v1/c3_math/al_curve.csv +7155ecbb5af764aff597a5071312c4fbe7ea3b9678d438e53a84db5994d544c3 results/historical-al-curves/v1/c4_code/al_curve.csv +7a9502461a40264469134c1602dce11605d746fb362d7946826addfabe04ee9c results/historical-al-curves/v1/c4_creative_writing/al_curve.csv +deec1ade9fbb0ca6120706265aebad6ed2ac5dfd1303a4235276dfe4dd86af9a results/historical-al-curves/v1/c4_factual_qa/al_curve.csv +e92a4cdf7a876ca2100795d03735b95741af6acd8346c11fb2baca421f9b8b13 results/historical-al-curves/v1/c4_general/al_curve.csv +e42b09abd0da17dadf2ec15a8f3a97ebd8abe052ed1cd28971fd35c4d82979b7 results/historical-al-curves/v1/c4_math/al_curve.csv +3b5f8477eb60ba034b09b80033824d38bc061efb987392bba95ebf82f84085b2 results/historical-al-curves/v1/c5_code/al_curve.csv +a7c05f9e0de662069c4c327511e5a44d8c5c2600a8a3d327d23a18e91f29fc75 results/historical-al-curves/v1/c5_creative_writing/al_curve.csv +e38286d943199cec158517b54a799287c37c50d78e996d30ec13c8099c5cb080 results/historical-al-curves/v1/c5_factual_qa/al_curve.csv +160423ad74f8f5bf5924b853e143138763795e3e7f0432729ebf07426d708c56 results/historical-al-curves/v1/c5_general/al_curve.csv +093d79e141db17e3f2da8411b754057585377e77dbe2ca4b0c021c18721233ce results/historical-al-curves/v1/c5_math/al_curve.csv +efa094eefd03080e2d1612e2780cabc4b4fe82102bae3a70d685bf1fd9a5bdce results/historical-al-curves/v1/c6_code/al_curve.csv +990c56cc487367e9e418bbeff9132144433f082632b7416e2d22dd36f003c689 results/historical-al-curves/v1/c6_creative_writing/al_curve.csv +e2ce13345d49e2960fa9e52649675e0a1c1af4b61f59e5d13a2375fb806263ee results/historical-al-curves/v1/c6_factual_qa/al_curve.csv +5d94f62ca87701b51af5ef930fe63be1df7958e85773ccd402edeacbf59a25b1 results/historical-al-curves/v1/c6_general/al_curve.csv +4fc731227fb3cc66f90a5152759c216faada25a65fc332d8148541b9f1c4bac0 results/historical-al-curves/v1/c6_math/al_curve.csv +4e8ea13685b72455b24ab172fa68d71743bc71a6d7ed7462fe30e36ee16a08d7 results/historical-al-curves/v1/c7_code/al_curve.csv +307586723ae39dda90995aded367c77aa254e3820a2ded9f42f2c62ef83f463f results/historical-al-curves/v1/c7_creative_writing/al_curve.csv +ecff8fdc0258485e27b1f4ebf11449d4345b6f12cae0c7ef4ecbd4c0d3705167 results/historical-al-curves/v1/c7_factual_qa/al_curve.csv +927daac26248cf56a8e7f09bd8c6b4a19a1f841cfb6b64d3d8643b5c0e017832 results/historical-al-curves/v1/c7_general/al_curve.csv +e4af3170b25fc05d0ad872adb322836852e15d0d40fca7b7fd4f7f7b27b8dd62 results/historical-al-curves/v1/c7_math/al_curve.csv +1789f2c326487f364f3c14bc9ef4773d2f5e6a3ddaee65cb8b1c2847f6116d47 results/historical-al-curves/v1/d0_code/al_curve.csv +cc4024858d44b0252ff081901cb2d710e83257679835f6d40ff8c8744db33602 results/historical-al-curves/v1/d0_code_bench/al_curve.csv +02c34be9c790947677503f07b3873849e14458c3cf18efed4717be917dadddab results/historical-al-curves/v1/d0_creative_writing/al_curve.csv +3eb6cfcd4fa4a755eb9d2f9a9c04ad4a82341a5eee734be5922d0a3530568c30 results/historical-al-curves/v1/d0_creative_writing_bench/al_curve.csv +2140012b6c2b0f52b2aada13cfad0cdae99afeb8b7936281c65e197c2f8b0fac results/historical-al-curves/v1/d0_factual_qa/al_curve.csv +b88004349fe0af1749dcfd00d3d1719c63079daab247bd961224ef7d5355eb20 results/historical-al-curves/v1/d0_factual_qa_bench/al_curve.csv +511471c17e2d15eedbd3481e7a9771170fccfdeeec09916974e3c1633793ee86 results/historical-al-curves/v1/d0_general/al_curve.csv +898934747a054fa24b146c14a5f2aba09d61152cc5c38a902cf4f56ff5af1d42 results/historical-al-curves/v1/d0_general_bench/al_curve.csv +5c4e2ae4caf4bef8303353cde38d98055123220e68f69c6896d65296edb81eb4 results/historical-al-curves/v1/d0_math/al_curve.csv +1e361705ccdbb2fa0507cfc3beffebddd392d36038efa3e8dc2542d281777545 results/historical-al-curves/v1/d0_math_bench/al_curve.csv +ebf1b0e2d4ec3da18de3011f029c029a81b8b14407c9b1d6a78333af3f07f433 results/historical-al-curves/v1/d1_code/al_curve.csv +8912e2738a3460457ee2b1968dd3dacf93e7fdaa6b22cc064359777f5f5ccf01 results/historical-al-curves/v1/d1_creative_writing/al_curve.csv +ed8eed33bff28b02c080d46b998217298a68e630abc23e732cc47251acadb2ab results/historical-al-curves/v1/d1_factual_qa/al_curve.csv +91824d146176f24973bfa6fbfe8a6d55249ac4881d1d5a23935c13f1d2c51618 results/historical-al-curves/v1/d1_general/al_curve.csv +8ccaf221637cc8cf7631ad7d9dd260adcf9f09520eff899ee72c5867f44f1714 results/historical-al-curves/v1/d1_math/al_curve.csv +2ba1138d82f05ed5fd7bc45186cff4351328324ad93608d2c25c20ddc8fe0b2d results/historical-al-curves/v1/d2_code/al_curve.csv +9c0dab3f0616fb762d43aeecab1e19d292d92ba3b72e74efc6ee3c37d2547dcf results/historical-al-curves/v1/d2_creative_writing/al_curve.csv +c63e517f6e5970d1b2208a455e50073b5fd531cfd9484b70890577a72baa74bf results/historical-al-curves/v1/d2_factual_qa/al_curve.csv +849b26bbb6d4cb5ec1b2ae207745d0c8cb5433b274404e116bb4a41bd9649f96 results/historical-al-curves/v1/d2_general/al_curve.csv +9a6dd3a37a3daf3368a2262863b82ae34e47192e25d16ab763b4dbd39f7a5799 results/historical-al-curves/v1/d2_math/al_curve.csv +48877d7e509598be3940beed6654af434efa1de2d9e4f46eb42352ddeb22c8ec results/historical-al-curves/v1/d3_code/al_curve.csv +2dc5e5d683438c79532dfabbe065e694a6b9603fa5ef7e73648eee7ad0ec7138 results/historical-al-curves/v1/d3_creative_writing/al_curve.csv +e2e5ed61a2ed83cd6e09030ce52692106a8ae7ebdbe1c18bdf58fc0abe150845 results/historical-al-curves/v1/d3_factual_qa/al_curve.csv +f080acc44e7d4ddc6a7cc253a722d9cfb73b531e235e0605101f951fed43e47c results/historical-al-curves/v1/d3_general/al_curve.csv +ba98a34a8360433870e3e12475fc275a7ab6d03bcc9b94029e7fa81a8dd97f2c results/historical-al-curves/v1/d3_math/al_curve.csv +06430bbb7a1dde00cb11eae287cf41212f0c4270fab85d6a7b4ea9cdfcc6376e results/historical-al-curves/v1/d4_code/al_curve.csv +138322d96b562a197c3dfce4be96877a63d5c60da9299baf9a4e1462e0457057 results/historical-al-curves/v1/d4_creative_writing/al_curve.csv +f6a4e0d1aabcd786e184e3845c2d8f07afff7938c8faacc6a6e5831024300e55 results/historical-al-curves/v1/d5_code/al_curve.csv +544e7047314513d17b726f10c4a47e3069d3d48eddb779cde9bef3f3aa9acb98 results/historical-al-curves/v1/d5_creative_writing/al_curve.csv +2c4c0da28aec12ee646a01c36e54e226f6f2f7c2598daa589e24083b753178dc results/historical-al-curves/v1/d5_factual_qa/al_curve.csv +db3224fa3489c952487463cb049c202e60a01ef6665d2d5a2aecc01d44c2985b results/historical-al-curves/v1/d5_general/al_curve.csv +e492b9a6a7e2fa83288c02c0f450739f2d30d124d6e05aa8cce23225792fdafe results/historical-al-curves/v1/d5_math/al_curve.csv +d6fe8d0beec5432eab9c8fe8a9b4fcbb7b08effc638f07bc8d56dd73cf18b1bb results/historical-al-curves/v1/d6_code/al_curve.csv +dcc8681fc1864c9af927f1acb89ca6e18cf2b8bb16738eb500779cdd79ef1321 results/historical-al-curves/v1/d6_creative_writing/al_curve.csv +ec1c8abcab0e73ceb72b7aeeaa56cacc483e032fe97bdb27b13096f659f933a4 results/historical-al-curves/v1/d6_factual_qa/al_curve.csv +38416012f7b76ee350ed97bb6efe3f2b3454a430a1a588e3a1007035804f26fc results/historical-al-curves/v1/d6_general/al_curve.csv +a9ed8256345309365b7cd216a0c587ab31c6e66d76eb9790c4b66dafd2e96859 results/historical-al-curves/v1/d6_math/al_curve.csv +cee9296d23831df5ee2411bb8a5f95c905796418fe20293c1a7c92981cdb99fe results/historical-al-curves/v1/d7_code/al_curve.csv +1dc7a8e5e0b04f920984b60ff2203bcf6dac7f48a5f331ffe0189bd893387d74 results/historical-al-curves/v1/d7_creative_writing/al_curve.csv +701dbf445877534d1817c9a7132023654f10ea50e4878c3aebd98a79614e60f5 results/historical-al-curves/v1/d7_factual_qa/al_curve.csv +bb45782350192168b00e368c3919d1a133096e2ba93b6c583f90ea903ab4a908 results/historical-al-curves/v1/d7_general/al_curve.csv +ca8299b3e74ed2a8b1831f7ecaf2cfecff9c8931f1396da632354b4b6e93a4d9 results/historical-al-curves/v1/d7_math/al_curve.csv +b5f3417a3a63c110dfa9d0bd7af497940873925f4f2ac969816034d64b54908c results/historical-al-curves/v1/gendense_l0/al_curve.csv +cc92b9d1cef97edf64e31bfabe6ef9b2dcfcc343d7f2c3fc492187adf72fce57 results/historical-al-curves/v1/gendense_l1/al_curve.csv +76d88265db58b5a24a79096ed0dc6dc8ca2e5e24024a15c8777e38e6e0895eee results/historical-al-curves/v1/gendense_l2/al_curve.csv +6e29784a0f479940051139ec7bc3b281c5d747f8169770daf008a05b64e72414 results/historical-al-curves/v1/gendense_l3/al_curve.csv +74378cdfcbbdb0342d963b8368350ca0a2dcaf7a9e331bbedd6b67bda9e313a5 results/historical-al-curves/v1/gendense_l4/al_curve.csv +cb8e712bf404f41ce40f3329276dcef71a9e05852fe4c230f2a21a23a4cf348c results/historical-al-curves/v1/gendense_l5/al_curve.csv +7cd3d94ff8987950768e4722014f3956f2af4114d1bdb428361028603b6d1088 results/historical-al-curves/v1/gendense_l6/al_curve.csv +e36b37efcd21b27b438bf1015266172c0804f369b48d64c243ab70cc9ab34867 results/historical-al-curves/v1/gendense_l7/al_curve.csv +edf093e7b7be16957278d8c961197b9e913be5b5ac0678df48bcd54e8c963fbf results/historical-al-curves/v1/genep4_code/al_curve.csv +644c907eb90c2ed4790bf0805df65af178fa4348076e6b7a5258bbe7bb5eb7e4 results/historical-al-curves/v1/genep4_math/al_curve.csv +f575b47ecd4d248d4f4c0f4a2e6e685ee60fc9aff335f0eae2e64d06d503ff17 results/historical-al-curves/v1/merge89/al_curve.csv +ffa68e561efb0ff335945d7f5ce2c4ce884c1cea35fbecc0acfc8f0a3a7f5a1c results/historical-al-curves/v1/mv2c89/al_curve.csv +acfb67b4fa59a0eb7a175f2c110e8b609eb2ff530283f5ad770e949b3d15d595 results/historical-al-curves/v1/mv3x89/al_curve.csv +62fe8604350ac825389c66aa3282b9c5076cfac2cbd33ce35f98222c8995edfd results/historical-al-curves/v1/mx_code_code/al_curve.csv +471ed1c8cb6266ce5462dd90f4fd767ff224b794b6431b51207b3db10e953d86 results/historical-al-curves/v1/mx_code_creative_writing/al_curve.csv +a0ac4e1243a51d3a5ab2aaf79595f8e89836f1f43210073c32fa36530bef3638 results/historical-al-curves/v1/mx_code_factual_qa/al_curve.csv +75ff837f1749d0f095ca7ba242cf6d00d7817b27049efa3c2cc4b51f9c5b4a9a results/historical-al-curves/v1/mx_code_general/al_curve.csv +27e2d44d0e1d64adbdc9dbb02f8fdfd5849a1acb65e886bca1b663dc947b1df2 results/historical-al-curves/v1/mx_code_math/al_curve.csv +193bbf84b8b4802b7ca15e36ff505558842da29023f3f450441b771e3244bdc0 results/historical-al-curves/v1/mx_creative_writing_code/al_curve.csv +0c193663d6bf65c6ffede829949cb80bc68801c4b52e7b3b254c410cb476e8d3 results/historical-al-curves/v1/mx_creative_writing_creative_writing/al_curve.csv +169dcde892e3f47236c2e5a440b18478aac40b6d578920e865124e3ce1ce729e results/historical-al-curves/v1/mx_creative_writing_factual_qa/al_curve.csv +79ffe500ee72208d201cde35dada407cd3120dd3761ebb53cc89107efac2420b results/historical-al-curves/v1/mx_creative_writing_general/al_curve.csv +4cb4168637b1d2ecfc71fe0709a3cabc8a59dbdf0d4a8894af7473afb3fe5458 results/historical-al-curves/v1/mx_creative_writing_math/al_curve.csv +0da911fc412c8831a08db40720644b2d07d1878e8b2d9505d2c344c58ed18f32 results/historical-al-curves/v1/mx_factual_qa_code/al_curve.csv +c22215e9aac307a3d72ec1dc4054e8c65bbfaae5e81c501d2e628b89e0ccac6a results/historical-al-curves/v1/mx_factual_qa_creative_writing/al_curve.csv +ebc0bf6bf21573185e2a4e494f4d2ba3a3fdefc0ab5bdcf7dd6cbb77bdb63e14 results/historical-al-curves/v1/mx_factual_qa_factual_qa/al_curve.csv +42edb75d5a4adece22f31fc5895568f5a0195373097fd356b1f463afaf03fca7 results/historical-al-curves/v1/mx_factual_qa_general/al_curve.csv +29dfae8c16ec355149bec331dab105867ec8741f0af6dc411d15d9074f2927ce results/historical-al-curves/v1/mx_factual_qa_math/al_curve.csv +ea26ed3625839230a90b0b5fb91ad8808856c13e16e1e303cf799c736c5b570d results/historical-al-curves/v1/mx_general_code/al_curve.csv +8e0fb5a0a6730b5f416190e30515cf294b4646c09d285f070d8d3eb15e0ce6bf results/historical-al-curves/v1/mx_general_creative_writing/al_curve.csv +d8767ce37119ecb19671fe82e28c340b6323d8c5fcce7e336ee004cc30472499 results/historical-al-curves/v1/mx_general_factual_qa/al_curve.csv +127bbd05b322c4e702ca8637c1754d299986f0ef8d08a9a37f8d734bb22e7ead results/historical-al-curves/v1/mx_general_general/al_curve.csv +3701966065391c5c89b1dca082fd9ab93ebc4ee444b51db0ae8d05268a5a50f1 results/historical-al-curves/v1/mx_general_math/al_curve.csv +a6b1afa6900cb541c4979a6714d42aed473565d9d646088961e2f5c4ccb6736b results/historical-al-curves/v1/mx_math_code/al_curve.csv +5eba966fd12aaf38b508164c325bb480d95aceec7a8fd3b359ff12084c8b8635 results/historical-al-curves/v1/mx_math_creative_writing/al_curve.csv +42c6c251cfe5c3559195b53a9860f00e545286f24aacf56e700f5d8c24c29aa8 results/historical-al-curves/v1/mx_math_factual_qa/al_curve.csv +0ef053311d4cae8ba0d339bf066e396bbc9624fccb00ee79a9ccf6215190f772 results/historical-al-curves/v1/mx_math_general/al_curve.csv +601ddb84962c2e9a12c49338d3688b534b1b6e8efc3ea1f7c47dfd8274ac3e85 results/historical-al-curves/v1/mx_math_math/al_curve.csv +ae5e4086fd51850db624e71aad1f6390564afea4c01909138b8ec518174a6ddd results/historical-al-curves/v1/rc89/al_curve.csv +1e682fc087dfa045ab6cd0604b6a6d60ccc9779188c383b3cce5d98922cb3348 results/historical-al-curves/v1/route89/al_curve.csv +226d571094d6326d5c77156c8b6f9b6f23565eb0a558589f23bc047a89b59fb9 results/historical-al-curves/v1/rw89x/al_curve.csv +d3786c58dd808b47532eadae27684b67369fd668725e3c58ab5b6d95fbb8eb0f results/historical-al-curves/v1/rwr2_0_code/al_curve.csv +98daad9455c1beef6de932d4c515c3f211ecce2b6db33d4195d2882f2a4ec8a0 results/historical-al-curves/v1/rwr2_1_code/al_curve.csv +ee1dc53951abdc88505e4eb04a160a2a0c069529ae4c7921f4a161c087bc7ed7 results/historical-al-curves/v1/rwr2_2_code/al_curve.csv +d66a15fbbacb240934dc919517d50be294ee76085c831d44adfc4131b9f9a324 results/historical-al-curves/v1/scnAg0_code/al_curve.csv +c0d060695f2f0c87fcd74cd065b720d47548eaf06d91dbc942fd1a8aa5d179ca results/historical-al-curves/v1/scnAg0_math/al_curve.csv +071f37849a80671bebc47c9c9bd9020b0a150dc39185575f323621cf5808e6a2 results/historical-al-curves/v1/scnAg1_code/al_curve.csv +f74a58358baddb9ad7038e3be19273d990e93fd1f6814d2c14b2fe9fe2a840ab results/historical-al-curves/v1/scnAg1_math/al_curve.csv +e81f887a2721c5e0334c956f7a4de7f31102b5a76b453e8510c34883172148d3 results/historical-al-curves/v1/scnAg2_code/al_curve.csv +b8d5de6fafb9437dd34f8dd0a92b861e3ff4303e21d37fdd82ea0ba426563025 results/historical-al-curves/v1/scnAg2_math/al_curve.csv +d177d4bd11c69bb5a669851dd02451a604ebeceb9d9763594d5f9f083bf75c85 results/historical-al-curves/v1/scnAg3_code/al_curve.csv +81a5d982e68efcd144f0353c18f97d1eb5372018d1e7bcdd0ebff652b15e16f7 results/historical-al-curves/v1/scnAg3_math/al_curve.csv +7de24f2594949bfe5ee6faa5ca99d976d72e97741ea961315aebfe1e447bd0d8 results/historical-al-curves/v1/scnAm0_math/al_curve.csv +3cbab716c23dd154accda87f4bcf794b2fc0651752aad17d823dc39b062b1660 results/historical-al-curves/v1/scnAm1_math/al_curve.csv +bf5646e0311035d6a58c3aa05f4c16c5a03ec9cad2a5e3f238d0870a468c932f results/historical-al-curves/v1/scnAm2_math/al_curve.csv +d1d8c2c472413573c1649ed6ece2cf45351f0f2801b90d3c1c3c1765de330ad0 results/historical-al-curves/v1/scnAm3_math/al_curve.csv +a27520a897853df3e676180297494c22b731734ac6b01dd7128ab0e28744dcae results/historical-al-curves/v1/scnAm_code/al_curve.csv +df9636a07e3c7c635e4c25e7c4c2ec39a062c3b691ed01d7630c902de36665cf results/historical-al-curves/v1/sd0_code/al_curve.csv +50f86a08650e19fdec4c1705af72c2667a25ad12c21fd228f540dfc17807fd91 results/historical-al-curves/v1/sd1_code/al_curve.csv +b966d37e18be328c18ac4488c53ca7ad314680d6878ba721b1f0b126bfde4351 results/historical-al-curves/v1/sd2_code/al_curve.csv +595ca5c6d0f9c564c9c4d3c1858259f14ba43a6c59a2c3a980017c8dddea6d5c results/historical-al-curves/v1/sd3_code/al_curve.csv +49f06e20a64540e35a1882f0d6f77c2ef97b0de40221adbdd4ce5ebb2e4fa6cc results/historical-al-curves/v1/sd4_code/al_curve.csv +203fe4ec71adef91a5e2c932d23b23a55f16b4db72cf8c8c26c10391807247f3 results/historical-al-curves/v1/sd5_code/al_curve.csv +93e9532b8afa55c7d04ebf21a097f5050719ae7faaf79c514a0b49fdb76433dd results/historical-al-curves/v1/sd5pk_creative_writing/al_curve.csv +6b48f906b7b3fc0ec47d1b4cb5a758c6779bae518405ab96773a31add496885e results/historical-al-curves/v1/sd5pk_factual_qa/al_curve.csv +b024f11ac64c058c715cb5652556058e912dc5b2b27e6a45f1c2a98ff9839b20 results/historical-al-curves/v1/sd5pk_general/al_curve.csv +86a5dc6aa38a11753c1f7c543bac0be7e9508412fd9d3621b8a1dccf675a52b7 results/historical-al-curves/v1/sd5pk_math/al_curve.csv +bd8c6ca5ee8bbc993cceb689528c6324f0c06175684ade3addd753d4e30bd124 results/historical-al-curves/v1/sd6_code/al_curve.csv +60e6b8260e89c9efab753e867c2bde1b4a8ee54b1715a7215216849297f33fb8 results/historical-al-curves/v1/sd7_code/al_curve.csv +e803f1bd631885a062f8cf78cfc78fd6010d5a75dd7a6575fc71256d386a5d91 results/historical-al-curves/v1/sg0_code/al_curve.csv +99e0a6ef3217f4d18ddc84092678467918f548c52e2fac892f779dade359cad5 results/historical-al-curves/v1/sg1_code/al_curve.csv +2312de5495b1bb06fd039a8ca49124b2ea19ae82174889c996aba3af5890cba7 results/historical-al-curves/v1/sg2_code/al_curve.csv +3abbfc13e520c00535e6cd2ecff9aa5aa91b8e3e4f1feca133a8c4557c580dbb results/historical-al-curves/v1/sg3_code/al_curve.csv +86a4d3bdeca1c89716a4afc24dc1f9a56a14e8df479fc8071fca91f8b0b3da05 results/historical-al-curves/v1/sg4_code/al_curve.csv +e5b0bec024da1370030d6c903dc14a353aa22d92a6c1a1efe2116374307265a8 results/historical-al-curves/v1/sg5_code/al_curve.csv +182709212d475a24728d84a642a0f589d0ad3807183a40a5375a95998f9ab05b results/historical-al-curves/v1/sg6_code/al_curve.csv +171d83662652e87ba1819c8e3ed696d30ae974994d486f50904a8673155c15a4 results/historical-al-curves/v1/sg7_code/al_curve.csv +763cb159181b98cffc15de07014e37bf2c378c84953a02147d860bfb76ff9f64 results/historical-al-curves/v1/sgpk_creative_writing/al_curve.csv +94d277ef87c5eb724a44e1e4e6e4bd310c2e86415ec10069c4cd65d111baf549 results/historical-al-curves/v1/sgpk_factual_qa/al_curve.csv +d9d8404db125c2cf0594ec8f30cc2f95fdf144eb2768dd1518dad1ad251bd08f results/historical-al-curves/v1/sgpk_general/al_curve.csv +566138ead9cd5d33e7434ac677ef7064c85c90587017164f9b12c89942b9f2d5 results/historical-al-curves/v1/sgpk_math/al_curve.csv +b21b54dcc70720f4b52af10e8d85653072580c197b8020c128871e208a462f9f results/historical-al-curves/v1/wr2be0_code/al_curve.csv +79a52b7a1a9ad1aa7412ec94606f1e762c1b5ede13fa3ea5c2a6d6da6509200f results/historical-al-curves/v1/wr2be1_code/al_curve.csv +b8c3d0f74241fd61dc13cb394b7780c546160e26730a0956d1938285f096a65e results/historical-al-curves/v1/wr2be2_code/al_curve.csv +fe39d6da277ea0f5fcaa73778bb46c2ef01a8309487153963f92764d1dc8d780 results/historical-al-curves/v1/wr3e0_code/al_curve.csv +dec31dbef3209b9019a41fe89f9c6a586f4a3dc0878287df741953101288a583 results/historical-al-curves/v1/wr3e1_code/al_curve.csv +6cfca51dd583996e8e68b1f2659aa472c2445ab4e0a847ef4d6a796276ae16fd results/historical-al-curves/v1/wr3e2_code/al_curve.csv +7e858ff1246214c81cf70976c920436d3271f9c77bd4e13a659ec9f4d3768864 results/main-figure-r1/v1/frozen-aggregate-cells.json +b2b335f41aa9a2381726c9d2c90aa8eb12a1a3b4f31ebfa60c31045ec168855d verification/v1/bootstrap_b1_exact_3p2m_routed.py +aee31b4fac56dba2507403e989daf7010a50a01fde3edb7ebca0bad57a90f2ae verification/v1/bootstrap_b2_epoch5_router.py +a20de477a77313a08619b1977b80e7fb7ffe83bd08bd923e0aa5c8132dd6901d verification/v1/bootstrap_paired_al.py +9734fc9b2b24a74ffa63708ae0ae7442962cef40f9147c14c3528c416c78d0ca verification/v1/r1_mainfig_verify_exports.py +f52bef510022d3ca470104bb184cdffe83ead49648dd712e64b0ea9149d901e3 verification/v1/verify_b1_exact_3p2m_routed_artifacts.py +7427838713af5748fbfd600af2b8a8d4d50d58a7403f072c352c025b12a0e3d1 verification/v1/verify_b2_epoch5_artifacts.py +67e2df392b3f69b5e9e8d4c0c2dddc20f7e5b79a9dcbb7f8cd3fffd278cf5a33 verification/v1/verify_b5_qwen3_4b_evidence.py diff --git a/docs/extended-evidence-v1/EXCLUSIONS.md b/docs/extended-evidence-v1/EXCLUSIONS.md new file mode 100644 index 0000000000000000000000000000000000000000..cd39063b72319371814d62a546d5e170bae60a99 --- /dev/null +++ b/docs/extended-evidence-v1/EXCLUSIONS.md @@ -0,0 +1,18 @@ +# Deliberate exclusions + +This bundle excludes: + +- `REPORT.md`, `EVIDENCE_LEDGER.md`, manuscript sources, and private planning; +- raw prompt datasets, prompt text, per-prompt generations, and router groups; +- raw benchmark/training logs and JSONL result sidecars; +- model weights, optimizer state, activations, prepared datasets, and caches; +- private internal training code or data; +- files containing credentials, usernames, or absolute internal paths; +- path-heavy R1/B1/C1/C3 manifests except the minimized aggregate R1 projection; +- `gen_curves.py`, `r1_mainfig_preflight.py`, + `r1_mainfig_validate_cells.py`, and `verify_c3_qwen3_4b_intake.py`, whose + current forms retain internal execution topology and need a separate + parameterization review. + +Historical aggregate curves and figures are retained for reuse, but their +presence does not promote them to current-paper evidence. diff --git a/docs/extended-evidence-v1/README.md b/docs/extended-evidence-v1/README.md new file mode 100644 index 0000000000000000000000000000000000000000..6417931905ac02d4c3a4cdda9cab3661196a20a1 --- /dev/null +++ b/docs/extended-evidence-v1/README.md @@ -0,0 +1,61 @@ +# MoS-DFlash extended evidence bundle v1 + +Frozen: `2026-07-23T20:26:58Z` + +This public-safe bundle contains reusable aggregate acceptance-length curves, +figure data, author-generated rendered figures, plotting recipes, and +verification utilities from the MoS-DFlash research workspace. + +## Evidence boundary + +- `results/historical-al-curves/v1/` contains aggregate AL values only. The + legacy `ckpt` column is a logical run/checkpoint identifier, not a filesystem + path. These curves are historical experiment records and are **not all + current-paper claims**. +- `results/main-figure-r1/v1/frozen-aggregate-cells.json` is a deliberately + minimized public projection. It retains only the complete 52/52 status and + the numeric panels consumed by the plotting code. All cell-level file paths, + prompt references, sidecars, and raw records were omitted. +- `figures/rendered/historical/v1/` preserves historical author-generated + figures. `figures/rendered/paper-current-candidate/v1/` records the selected + paper assets at this freeze; the manuscript remains the authority for which + figures are finally submitted. +- `recipes/plotting/v1/` and `verification/v1/` contain code, not raw inputs. + Callers must supply their own licensed datasets, model paths, and aggregate + sidecars where required. + +## Reproducible examples + +```bash +python recipes/plotting/v1/plot_main_results.py \ + --evidence results/main-figure-r1/v1/frozen-aggregate-cells.json \ + --output /tmp/fig_main_results.png + +python recipes/plotting/v1/plot_mos_5x5_gains.py \ + --evidence results/main-figure-r1/v1/frozen-aggregate-cells.json \ + --output-dir /tmp/mos-matrices + +python recipes/plotting/v1/build_mos_architecture_drawio.py \ + --output /tmp/fig_architecture_v8.drawio +``` + +The Qwen3-4B matched-volume plotting recipe expects the aggregate trajectory +from the separately frozen `releases/b5-qwen3-4b-fixed-budget/` evidence lane. + +## Audit + +- `manifests/extended-evidence-v1/artifact-manifest.jsonl` maps every published + content object to its repository-relative source path and source SHA-256. +- `checksums/extended-evidence-v1.sha256` freezes staged bytes. +- Files with hard-coded Weka or `/tmp` output locations were not copied as-is. + Only explicit sanitized copies with relative output paths are present, and + their source and staged hashes differ in the manifest. + +No raw prompts, per-prompt generations, benchmark/training logs, private model +or activation material, credentials, absolute internal paths, `REPORT.md`, or +`EVIDENCE_LEDGER.md` are included. + +## License + +Code and author-generated artifacts in this bundle are released under the MIT +license in `licenses/extended-evidence-v1/LICENSE`. diff --git a/figures/data/qwen3-4b/v1/q3_4b_armA_traj.csv b/figures/data/qwen3-4b/v1/q3_4b_armA_traj.csv new file mode 100644 index 0000000000000000000000000000000000000000..fe418bbb7800232ab2ced2625697075e25a7c467 --- /dev/null +++ b/figures/data/qwen3-4b/v1/q3_4b_armA_traj.csv @@ -0,0 +1,33 @@ +point,step,overall_al,code,math,factual_qa,creative_writing,general +step_10000,10000,3.1315,3.1009,4.4019,2.6961,2.6091,2.8495 +step_12500,12500,3.1969,3.1276,4.5913,2.7403,2.6474,2.8777 +step_15000,15000,3.2012,3.1329,4.5902,2.7227,2.6764,2.8840 +step_17500,17500,3.2312,3.1796,4.5761,2.7718,2.6783,2.9503 +step_20000,20000,3.2405,3.1978,4.6114,2.7855,2.6783,2.9295 +step_22500,22500,3.2702,3.2066,4.6825,2.7797,2.7210,2.9610 +step_2500,2500,2.9335,2.8528,4.1414,2.5529,2.3996,2.7209 +step_25000,25000,3.2897,3.2198,4.7660,2.8254,2.6797,2.9575 +step_27500,27500,3.2950,3.2186,4.7407,2.7984,2.7394,2.9779 +step_30000,30000,3.3085,3.2391,4.7723,2.8213,2.7316,2.9781 +step_32500,32500,3.3256,3.2592,4.8172,2.8294,2.7085,3.0135 +step_35000,35000,3.3261,3.2914,4.8069,2.8414,2.6962,2.9948 +step_37500,37500,3.3239,3.2841,4.8004,2.8284,2.7176,2.9890 +step_40000,40000,3.3506,3.2863,4.8107,2.8388,2.7968,3.0202 +step_42500,42500,3.3679,3.2620,4.9353,2.8437,2.7576,3.0407 +step_45000,45000,3.3627,3.2946,4.8828,2.8456,2.7726,3.0178 +step_47500,47500,3.3761,3.3020,4.8511,2.8643,2.8334,3.0299 +step_5000,5000,3.0275,2.9729,4.2464,2.6362,2.5029,2.7791 +step_50000,50000,3.3793,3.3113,4.9204,2.8589,2.7821,3.0235 +step_52500,52500,3.3732,3.3012,4.9109,2.8472,2.7816,3.0251 +step_55000,55000,3.3787,3.3068,4.8855,2.8820,2.7892,3.0299 +step_57500,57500,3.3867,3.3318,4.8669,2.8990,2.8080,3.0275 +step_60000,60000,3.3851,3.3215,4.9434,2.8583,2.7836,3.0184 +step_62500,62500,3.3869,3.2986,4.9339,2.8783,2.8065,3.0170 +step_65000,65000,3.3844,3.3203,4.8788,2.8800,2.7953,3.0476 +step_67500,67500,3.3854,3.3001,4.8815,2.8600,2.8460,3.0393 +step_70000,70000,3.3894,3.3288,4.9029,2.8620,2.8070,3.0465 +step_72500,72500,3.3713,3.3143,4.8882,2.8801,2.7253,3.0484 +step_7500,7500,3.0940,3.0365,4.3622,2.6655,2.5707,2.8351 +epoch_0_step_24970,24970,3.2848,3.2287,4.6862,2.8052,2.7273,2.9769 +epoch_1_step_49940,49940,3.3640,3.2700,4.8682,2.8645,2.8029,3.0143 +epoch_2_step_74910,74910,3.3845,3.3085,4.9530,2.8821,2.7355,3.0432 diff --git a/figures/data/qwen3-4b/v1/q3_4b_armB_traj.csv b/figures/data/qwen3-4b/v1/q3_4b_armB_traj.csv new file mode 100644 index 0000000000000000000000000000000000000000..fad24e5bb84db372ee8064a214e8a1162f734be7 --- /dev/null +++ b/figures/data/qwen3-4b/v1/q3_4b_armB_traj.csv @@ -0,0 +1,33 @@ +point,step,overall_al,code,math,factual_qa,creative_writing,general +step_10000,10000,3.3373,3.2303,4.7736,2.8494,2.7661,3.0674 +step_12500,12500,3.3604,3.2559,4.8735,2.8494,2.7591,3.0638 +step_15000,15000,3.3424,3.2609,4.7799,2.8671,2.7336,3.0708 +step_17500,17500,3.3587,3.2884,4.7953,2.8823,2.7616,3.0662 +step_20000,20000,3.3828,3.2908,4.8577,2.8816,2.7912,3.0929 +step_22500,22500,3.4117,3.3280,4.8962,2.8947,2.8319,3.1077 +step_2500,2500,3.2904,3.2040,4.6521,2.8480,2.7205,3.0272 +step_25000,25000,3.4146,3.3412,4.9544,2.8993,2.7676,3.1106 +step_27500,27500,3.3808,3.3244,4.8682,2.8768,2.7483,3.0864 +step_30000,30000,3.4306,3.3264,4.9819,2.9117,2.8277,3.1051 +step_32500,32500,3.4111,3.3200,4.9407,2.8812,2.8235,3.0901 +step_35000,35000,3.4409,3.3530,4.9654,2.9262,2.8673,3.0927 +step_37500,37500,3.4415,3.3442,4.9805,2.9096,2.8555,3.1174 +step_40000,40000,3.4502,3.3550,5.0112,2.9184,2.8487,3.1178 +step_42500,42500,3.4495,3.3484,5.0098,2.9469,2.8272,3.1151 +step_45000,45000,3.4678,3.3501,5.0866,2.9109,2.8646,3.1270 +step_47500,47500,3.4720,3.3631,5.0507,2.9503,2.8834,3.1122 +step_5000,5000,3.3201,3.2545,4.7445,2.8371,2.7114,3.0531 +step_50000,50000,3.4360,3.3625,4.9667,2.9192,2.8004,3.1313 +step_52500,52500,3.4759,3.3722,5.0564,2.9192,2.9047,3.1270 +step_55000,55000,3.4906,3.3837,5.0837,2.9562,2.8976,3.1320 +step_57500,57500,3.4824,3.3605,5.1011,2.9401,2.8743,3.1363 +step_60000,60000,3.4746,3.3761,5.0779,2.9522,2.8455,3.1212 +step_62500,62500,3.4826,3.3732,5.0982,2.9484,2.8856,3.1076 +step_65000,65000,3.4856,3.3630,5.1509,2.9388,2.8593,3.1162 +step_67500,67500,3.4704,3.3592,5.1142,2.9123,2.8272,3.1390 +step_70000,70000,3.4722,3.3882,5.0450,2.9536,2.8582,3.1158 +step_72500,72500,3.4776,3.3580,5.1156,2.9522,2.8587,3.1035 +step_7500,7500,3.3310,3.2471,4.7096,2.8579,2.7861,3.0542 +epoch_0_step_24970,24970,3.3997,3.3121,4.9177,2.9156,2.7651,3.0880 +epoch_1_step_49940,49940,3.4776,3.3692,5.0779,2.9344,2.8970,3.1094 +epoch_2_step_74910,74910,3.4775,3.3702,5.0536,2.9517,2.8796,3.1326 diff --git a/figures/data/qwen3-4b/v1/q3_4b_gen_traj.csv b/figures/data/qwen3-4b/v1/q3_4b_gen_traj.csv new file mode 100644 index 0000000000000000000000000000000000000000..77a65a622cd5e4b6abcf06db0c95fecf3c7c5305 --- /dev/null +++ b/figures/data/qwen3-4b/v1/q3_4b_gen_traj.csv @@ -0,0 +1,30 @@ +step,overall_al,code,math,factual_qa,creative_writing,general +10000,3.1686,3.0954,4.4225,2.7484,2.6225,2.9544 +11250,3.1772,3.1039,4.4655,2.7516,2.6328,2.9324 +1250,2.8705,2.8119,3.8965,2.5357,2.4068,2.7016 +12500,3.2067,3.1265,4.5161,2.7495,2.6620,2.9796 +13750,3.2080,3.1213,4.5298,2.7881,2.6315,2.9692 +15000,3.2329,3.1571,4.5298,2.7828,2.6881,3.0066 +16250,3.2303,3.1638,4.5333,2.8110,2.6524,2.9912 +17500,3.2244,3.1572,4.5287,2.8225,2.6216,2.9920 +18750,3.2505,3.1740,4.5890,2.8148,2.6895,2.9849 +20000,3.2580,3.1773,4.6245,2.8278,2.6648,2.9958 +21250,3.2841,3.2001,4.6594,2.8040,2.7143,3.0429 +22500,3.2748,3.2091,4.6305,2.8277,2.6620,3.0446 +23750,3.2820,3.2021,4.6091,2.8431,2.7181,3.0378 +2500,2.9964,2.9148,4.1491,2.6310,2.5090,2.7783 +25000,3.3024,3.2186,4.6353,2.8531,2.7463,3.0587 +26250,3.2875,3.2030,4.6233,2.8247,2.7414,3.0452 +27500,3.2927,3.2635,4.6114,2.8239,2.7157,3.0490 +28750,3.2976,3.2152,4.6911,2.8471,2.7057,3.0287 +30000,3.2872,3.2221,4.6485,2.8589,2.6593,3.0472 +31250,3.2953,3.2416,4.6257,2.8402,2.6999,3.0691 +32500,3.2755,3.1940,4.6257,2.8419,2.6853,3.0304 +33750,3.3071,3.2128,4.6789,2.8465,2.7220,3.0751 +35000,3.2808,3.2014,4.6642,2.8179,2.7018,3.0183 +36250,3.2694,3.1896,4.6233,2.8381,2.6616,3.0345 +3750,3.0412,2.9734,4.2085,2.6458,2.5334,2.8447 +5000,3.0631,3.0204,4.2495,2.6940,2.5024,2.8490 +6250,3.1019,3.0239,4.3264,2.6841,2.5902,2.8851 +7500,3.1331,3.0864,4.3506,2.7112,2.5946,2.9230 +8750,3.1352,3.0659,4.3782,2.7174,2.5937,2.9209 diff --git a/figures/rendered/historical/v1/code250k_vs_gen.png b/figures/rendered/historical/v1/code250k_vs_gen.png new file mode 100644 index 0000000000000000000000000000000000000000..58ea93cb4dec2e0532c936c5fdb6debe1a5a5412 --- /dev/null +++ b/figures/rendered/historical/v1/code250k_vs_gen.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89992f95494655a1263b1f17c098c97e68bba6aa19279f29b7d22a049096ce50 +size 77225 diff --git a/figures/rendered/historical/v1/fig_bigdata_vs_gen.png b/figures/rendered/historical/v1/fig_bigdata_vs_gen.png new file mode 100644 index 0000000000000000000000000000000000000000..0a51df1b63999a67e2e502962a6e1a74e6c066be --- /dev/null +++ b/figures/rendered/historical/v1/fig_bigdata_vs_gen.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:66e0cfc000e9583110ce5d174ab1874210eee0b026d25fd863e7f69139f78ec3 +size 74481 diff --git a/figures/rendered/historical/v1/fig_exp1_forgetting.png b/figures/rendered/historical/v1/fig_exp1_forgetting.png new file mode 100644 index 0000000000000000000000000000000000000000..f192863be1cbcb4cc614f8015251d1f4b67e7a7b --- /dev/null +++ b/figures/rendered/historical/v1/fig_exp1_forgetting.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c88b5a91d84cfbf30f5f9b1813ae470f808e348d27f224f57b8704490f839016 +size 347513 diff --git a/figures/rendered/historical/v1/fig_exp1_three_recipes.png b/figures/rendered/historical/v1/fig_exp1_three_recipes.png new file mode 100644 index 0000000000000000000000000000000000000000..733f947fc4463c9a33e40d8d2e808d2b7579a60c --- /dev/null +++ b/figures/rendered/historical/v1/fig_exp1_three_recipes.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b9633ec3524f23766a72e788206c114e02d7b0da5be8f3cf8a501848fe2fb52 +size 80574 diff --git a/figures/rendered/historical/v1/fig_exp4_inference.png b/figures/rendered/historical/v1/fig_exp4_inference.png new file mode 100644 index 0000000000000000000000000000000000000000..3353e8e67dc917abf6776f669ae73eaa934889d5 --- /dev/null +++ b/figures/rendered/historical/v1/fig_exp4_inference.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8602bc3791560060deba3ad715eedf6fd9d1ef5d76c1774d3ef5a5fb0ccb2bd +size 147109 diff --git a/figures/rendered/historical/v1/fig_exp5_method_vs_data.png b/figures/rendered/historical/v1/fig_exp5_method_vs_data.png new file mode 100644 index 0000000000000000000000000000000000000000..e5833fbb40088fbf7e16ed49ab40c70c061a2c2b --- /dev/null +++ b/figures/rendered/historical/v1/fig_exp5_method_vs_data.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf243131b8c61f12c258760307178a8d7888e41c5ef1f5a8981f37b7f639b83a +size 108434 diff --git a/figures/rendered/historical/v1/fig_forgetting_matrix.png b/figures/rendered/historical/v1/fig_forgetting_matrix.png new file mode 100644 index 0000000000000000000000000000000000000000..b593ea122667b273300e11bb6a8b284f337121f1 --- /dev/null +++ b/figures/rendered/historical/v1/fig_forgetting_matrix.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58be85cb2527aa44c3e8c2579831138104bc70fa7e33794e1855df76988e9d6d +size 133775 diff --git a/figures/rendered/historical/v1/fig_fusion_combined.png b/figures/rendered/historical/v1/fig_fusion_combined.png new file mode 100644 index 0000000000000000000000000000000000000000..e12aaf4ed038dfc65c48db9d300d3c55a85f0102 --- /dev/null +++ b/figures/rendered/historical/v1/fig_fusion_combined.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:73c7ceff7bf8f01c0bc3e861c689532619dc2f358f51b37a0d9cedd8b07779bc +size 98630 diff --git a/figures/rendered/historical/v1/fig_gen800k_loss_al.png b/figures/rendered/historical/v1/fig_gen800k_loss_al.png new file mode 100644 index 0000000000000000000000000000000000000000..80419cd9a80dc676e98c31e64b97288b660d8b73 --- /dev/null +++ b/figures/rendered/historical/v1/fig_gen800k_loss_al.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc68719d712a0bc37321c46ec7d9b6368c08fc3f4523ef1d1c3d7771e5a99a37 +size 163009 diff --git a/figures/rendered/historical/v1/fig_merged_vs_specialist.png b/figures/rendered/historical/v1/fig_merged_vs_specialist.png new file mode 100644 index 0000000000000000000000000000000000000000..cf4cf149682c3e2e1106e7ff53fa38474b67177a --- /dev/null +++ b/figures/rendered/historical/v1/fig_merged_vs_specialist.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca381bfb16abfde13f5e88c917b27698155fa073e99cba70ad50235c6565a155 +size 99807 diff --git a/figures/rendered/historical/v1/fig_q3_4b_traj.png b/figures/rendered/historical/v1/fig_q3_4b_traj.png new file mode 100644 index 0000000000000000000000000000000000000000..250adfb694369cee7c8675fbdcb42432a1c0773f --- /dev/null +++ b/figures/rendered/historical/v1/fig_q3_4b_traj.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16825387c4de0251447e50ba54aa99585cf8fc2790b1d0f4c8438035c40e23f0 +size 109084 diff --git a/figures/rendered/historical/v1/fig_serving.png b/figures/rendered/historical/v1/fig_serving.png new file mode 100644 index 0000000000000000000000000000000000000000..b0d5394a925858d27d8912def4db14223f8db417 --- /dev/null +++ b/figures/rendered/historical/v1/fig_serving.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77ae4d2ff2f302212f1252b8bd1fbdc3c0d244dfd2a1be06d093a964575eff8e +size 107592 diff --git a/figures/rendered/historical/v1/fig_smalldata_vs_gen.png b/figures/rendered/historical/v1/fig_smalldata_vs_gen.png new file mode 100644 index 0000000000000000000000000000000000000000..e0ab84a242d64a98ca70f8fd1e18fa2dfde3866d --- /dev/null +++ b/figures/rendered/historical/v1/fig_smalldata_vs_gen.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b8928ad66e0c86b58bc8d9088af11647e78f4f42180e95a0a5bd96017f623e3 +size 73631 diff --git a/figures/rendered/historical/v1/fig_specialist_overall.png b/figures/rendered/historical/v1/fig_specialist_overall.png new file mode 100644 index 0000000000000000000000000000000000000000..f26a3a5f9fb5fe8cc486da54409ab07996286491 --- /dev/null +++ b/figures/rendered/historical/v1/fig_specialist_overall.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e3e6ae946db8f2922832f3b0b09673877c0cf68a55f9027fdc93d60399b9446 +size 57144 diff --git a/figures/rendered/historical/v1/fig_warm_saturation.png b/figures/rendered/historical/v1/fig_warm_saturation.png new file mode 100644 index 0000000000000000000000000000000000000000..35338070dba2ede2d6992f4871b3a72447230003 --- /dev/null +++ b/figures/rendered/historical/v1/fig_warm_saturation.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:861c7f202252b6798391fb5bc98bc88d7379e38b3c7d77d2e4fbf0f74f144085 +size 117399 diff --git a/figures/rendered/historical/v1/gen_per_domain_AL.png b/figures/rendered/historical/v1/gen_per_domain_AL.png new file mode 100644 index 0000000000000000000000000000000000000000..e7a5eb87c70c83bd4adcd5913da3da040985c408 --- /dev/null +++ b/figures/rendered/historical/v1/gen_per_domain_AL.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb2084ef355799b98ab08000994e54aa8966196cbb3485235a942ed8966ac32c +size 34837 diff --git a/figures/rendered/historical/v1/gen_vs_merge_attn.png b/figures/rendered/historical/v1/gen_vs_merge_attn.png new file mode 100644 index 0000000000000000000000000000000000000000..147b0a39d11e98120fd3993c560b4cb29b1e74ca --- /dev/null +++ b/figures/rendered/historical/v1/gen_vs_merge_attn.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f484b361bed4165c4aac24a52ad66b7a3f9ea7948d7f617c764784085ac39b5 +size 136072 diff --git a/figures/rendered/historical/v1/joint_vs_pipeline_code.png b/figures/rendered/historical/v1/joint_vs_pipeline_code.png new file mode 100644 index 0000000000000000000000000000000000000000..3b0fdc7f0142cd11f1646a0ecabed505da2da6e2 --- /dev/null +++ b/figures/rendered/historical/v1/joint_vs_pipeline_code.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:221d21d5085f280ace83fe2532a3ae52858a6787af5f28addfd480095114984b +size 172019 diff --git a/figures/rendered/historical/v1/overfit_attention_ladder.png b/figures/rendered/historical/v1/overfit_attention_ladder.png new file mode 100644 index 0000000000000000000000000000000000000000..16ba91d96a0cc7718508ce7be8f4b22c4093add2 --- /dev/null +++ b/figures/rendered/historical/v1/overfit_attention_ladder.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7bb2f21888c8c79e731add5c61021aff4be2c58a9dfc499601d902700867802c +size 222223 diff --git a/figures/rendered/historical/v1/overfit_loss_al.png b/figures/rendered/historical/v1/overfit_loss_al.png new file mode 100644 index 0000000000000000000000000000000000000000..1ad28915727615b0f8d297916238eeb3c805465b --- /dev/null +++ b/figures/rendered/historical/v1/overfit_loss_al.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8b24897fdce78a542aa0f9387e13299ffe3a34c679b607470ac984bddbbf10c +size 181649 diff --git a/figures/rendered/historical/v1/reasonmix_5arm_per_domain.png b/figures/rendered/historical/v1/reasonmix_5arm_per_domain.png new file mode 100644 index 0000000000000000000000000000000000000000..49efdd9f4a9ab5c87df02f9c7f35fee78f13105f --- /dev/null +++ b/figures/rendered/historical/v1/reasonmix_5arm_per_domain.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77b4ad8a26a01b67b052d7374de6b11350d89ba6eace1e8d135af159ffca5c7f +size 96903 diff --git a/figures/rendered/historical/v1/recipe_budget.png b/figures/rendered/historical/v1/recipe_budget.png new file mode 100644 index 0000000000000000000000000000000000000000..7cd8cf42978fb343f8a6ed706b375f18a0b4760f --- /dev/null +++ b/figures/rendered/historical/v1/recipe_budget.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c314a567c8cadfcfb602bc0ad10c328b6dca571213b88ef656bf78404350af4 +size 359780 diff --git a/figures/rendered/paper-current-candidate/v1/fig_architecture_v8.drawio b/figures/rendered/paper-current-candidate/v1/fig_architecture_v8.drawio new file mode 100644 index 0000000000000000000000000000000000000000..ecf041a54ac714af27b53638898e02b28cc28b63 --- /dev/null +++ b/figures/rendered/paper-current-candidate/v1/fig_architecture_v8.drawio @@ -0,0 +1,728 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + \ No newline at end of file diff --git a/figures/rendered/paper-current-candidate/v1/fig_architecture_v8.png b/figures/rendered/paper-current-candidate/v1/fig_architecture_v8.png new file mode 100644 index 0000000000000000000000000000000000000000..952d3591060a93f1ed64a626a0de8ef797a99438 --- /dev/null +++ b/figures/rendered/paper-current-candidate/v1/fig_architecture_v8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3961c8a6ccc80992197b8746453cffd1d6f87ed158ce6103a72b4db361996f66 +size 1543988 diff --git a/figures/rendered/paper-current-candidate/v1/fig_architecture_v8.svg b/figures/rendered/paper-current-candidate/v1/fig_architecture_v8.svg new file mode 100644 index 0000000000000000000000000000000000000000..0b6d3e868c7a14ce70fed69d02bb6090e015ceb0 --- /dev/null +++ b/figures/rendered/paper-current-candidate/v1/fig_architecture_v8.svg @@ -0,0 +1,4 @@ + + + +AConstruct training signalsTRAINING GROUPSDomain 1Domain 2⋯Domain NRequest–responsepair (x, y)Prompt-onlytraining label dResponse yy1y2yay4⋯sampled anchorMasked candidate blockya[M][M][M]Frozen targetFrozen featuresTarget contextztPrompt featuresq(x)BJointly train MoSONE EXPANDED DRAFT LAYERhℓNormSharedattention+NormMLP bankMLPℓ,1MLPℓ,2MLPℓ,d⋯MLPℓ,N+hℓ+1zt1zt2⋯Target-contextprojectionTraining label dresidualresidualSame d across layersMLP1,dMLP2,dMLPL,d⋯InitializationPublic DFlash or trained generalist↓Shared modules +copied MLP groupsParallel predictionsBlock-prediction lossUpdates shared modulesand selected MLPROUTER SUPERVISIONPrompt features(no gradient)Mean poolRequest routerRouting lossRouter updateMoS checkpointShared modules · MLP bankRequest routerCRequest-routed speculative decodingROUTE ONCEPrompt xFrozen targetprompt passPrompt featuresMean poolRequest routerPredictedgroup dpredLOCKOne decision per requestcycle 1 — cycle 2 — ⋯ — cycle TFixed for all layers and cyclesTrained MoS checkpointDATA PATHTarget contextztMasked block[M] [M] [M]Selected MoS pathfixed group dpred for the whole requestLayer 1SharedmodulesMLP1Layer 2SharedmodulesMLP2Layer LSharedmodulesMLPLOther MLP groups are not evaluated or mergedCandidate block (parallel)t1t2t3⋯t16Target verificationt1t2t3t4✓ ✓ ✓fixAccept matching prefixCorrect first mismatchResponseAccepted or corrected tokens start the next verification cyclesharedselected MLProutingfrozen / inactiveExact verification preserves the target distribution \ No newline at end of file diff --git a/figures/rendered/paper-current-candidate/v1/fig_main_results.png b/figures/rendered/paper-current-candidate/v1/fig_main_results.png new file mode 100644 index 0000000000000000000000000000000000000000..73a2ec1f1c7bdb988f2f18ea59ca828f0471375c --- /dev/null +++ b/figures/rendered/paper-current-candidate/v1/fig_main_results.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc224065a687b7631cec8da84803ec6b8b1c5c34b737dfaa4e7077c9eb04318d +size 450639 diff --git a/figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.pdf b/figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.pdf new file mode 100644 index 0000000000000000000000000000000000000000..fad9b0749aea4efa465deef31a9d398a46fdc776 Binary files /dev/null and b/figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.pdf differ diff --git a/figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.png b/figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.png new file mode 100644 index 0000000000000000000000000000000000000000..46191c8057107ca9da8063a0de48ae55aa230f75 --- /dev/null +++ b/figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1243e3db541c0be5dbdacdabf6046ad065911ec2080bacd7d6e69bc9100689db +size 230257 diff --git a/licenses/extended-evidence-v1/LICENSE b/licenses/extended-evidence-v1/LICENSE new file mode 100644 index 0000000000000000000000000000000000000000..909b8ff34ce3ff391ec5ecd1d2388d0c5b1cd4b3 --- /dev/null +++ b/licenses/extended-evidence-v1/LICENSE @@ -0,0 +1,21 @@ +MIT License + +Copyright (c) 2025 sgl-project + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. diff --git a/manifests/extended-evidence-v1/artifact-manifest.jsonl b/manifests/extended-evidence-v1/artifact-manifest.jsonl new file mode 100644 index 0000000000000000000000000000000000000000..bad5d41befa5e6dfc314550c71e14f85df60a04f --- /dev/null +++ b/manifests/extended-evidence-v1/artifact-manifest.jsonl @@ -0,0 +1,308 @@ +{"artifact_type": "documentation", "license": "mit", "size_bytes": 888, "source_path": null, "source_sha256": null, "staged_path": "docs/extended-evidence-v1/EXCLUSIONS.md", "staged_sha256": "2469e607e691cba884beda0fe55fe69bf1dbb14ca8f22765a81139f6bf60d18d", "status": "current", "transformation": "generated-public-documentation"} +{"artifact_type": "documentation", "license": "mit", "size_bytes": 2695, "source_path": null, "source_sha256": null, "staged_path": "docs/extended-evidence-v1/README.md", "staged_sha256": "41d8ad21168e4958be51f2a6649c585c2151c807d95f6d05daaef0d837ee0404", "status": "current", "transformation": "generated-public-documentation"} +{"artifact_type": "figure-data", "license": "mit", "size_bytes": 1974, "source_path": "experiments/dflash/figs/src/q3_4b_armA_traj.csv", "source_sha256": "70082850ca060f1a2b0a7762c85ba07d515b4c66d87382870dda0be5a5ceafd1", "staged_path": "figures/data/qwen3-4b/v1/q3_4b_armA_traj.csv", "staged_sha256": "70082850ca060f1a2b0a7762c85ba07d515b4c66d87382870dda0be5a5ceafd1", "status": "frozen-aggregate", "transformation": "direct-copy"} +{"artifact_type": "figure-data", "license": "mit", "size_bytes": 1974, "source_path": "experiments/dflash/figs/src/q3_4b_armB_traj.csv", "source_sha256": "6f406c21eea7ee25bda9c00bec301d705a3427bf2de2e6ab839ef31cc701090a", "staged_path": "figures/data/qwen3-4b/v1/q3_4b_armB_traj.csv", "staged_sha256": "6f406c21eea7ee25bda9c00bec301d705a3427bf2de2e6ab839ef31cc701090a", "status": "frozen-aggregate", "transformation": "direct-copy"} +{"artifact_type": "figure-data", "license": "mit", "size_bytes": 1447, "source_path": "experiments/dflash/figs/src/q3_4b_gen_traj.csv", "source_sha256": "5dd37b6c7d9a3d4c6ede536aec2a869aa26afd8c67c3096a04e66355827815e9", "staged_path": "figures/data/qwen3-4b/v1/q3_4b_gen_traj.csv", "staged_sha256": "5dd37b6c7d9a3d4c6ede536aec2a869aa26afd8c67c3096a04e66355827815e9", "status": "frozen-aggregate", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 77225, "source_path": "experiments/dflash/figs/code250k_vs_gen.png", "source_sha256": "89992f95494655a1263b1f17c098c97e68bba6aa19279f29b7d22a049096ce50", "staged_path": "figures/rendered/historical/v1/code250k_vs_gen.png", "staged_sha256": "89992f95494655a1263b1f17c098c97e68bba6aa19279f29b7d22a049096ce50", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 74481, "source_path": "experiments/dflash/figs/fig_bigdata_vs_gen.png", "source_sha256": "66e0cfc000e9583110ce5d174ab1874210eee0b026d25fd863e7f69139f78ec3", "staged_path": "figures/rendered/historical/v1/fig_bigdata_vs_gen.png", "staged_sha256": "66e0cfc000e9583110ce5d174ab1874210eee0b026d25fd863e7f69139f78ec3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 347513, "source_path": "experiments/dflash/figs/fig_exp1_forgetting.png", "source_sha256": "c88b5a91d84cfbf30f5f9b1813ae470f808e348d27f224f57b8704490f839016", "staged_path": "figures/rendered/historical/v1/fig_exp1_forgetting.png", "staged_sha256": "c88b5a91d84cfbf30f5f9b1813ae470f808e348d27f224f57b8704490f839016", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 80574, "source_path": "experiments/dflash/figs/fig_exp1_three_recipes.png", "source_sha256": "2b9633ec3524f23766a72e788206c114e02d7b0da5be8f3cf8a501848fe2fb52", "staged_path": "figures/rendered/historical/v1/fig_exp1_three_recipes.png", "staged_sha256": "2b9633ec3524f23766a72e788206c114e02d7b0da5be8f3cf8a501848fe2fb52", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 147109, "source_path": "experiments/dflash/figs/fig_exp4_inference.png", "source_sha256": "e8602bc3791560060deba3ad715eedf6fd9d1ef5d76c1774d3ef5a5fb0ccb2bd", "staged_path": "figures/rendered/historical/v1/fig_exp4_inference.png", "staged_sha256": "e8602bc3791560060deba3ad715eedf6fd9d1ef5d76c1774d3ef5a5fb0ccb2bd", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 108434, "source_path": "experiments/dflash/figs/fig_exp5_method_vs_data.png", "source_sha256": "bf243131b8c61f12c258760307178a8d7888e41c5ef1f5a8981f37b7f639b83a", "staged_path": "figures/rendered/historical/v1/fig_exp5_method_vs_data.png", "staged_sha256": "bf243131b8c61f12c258760307178a8d7888e41c5ef1f5a8981f37b7f639b83a", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 133775, "source_path": "experiments/dflash/figs/fig_forgetting_matrix.png", "source_sha256": "58be85cb2527aa44c3e8c2579831138104bc70fa7e33794e1855df76988e9d6d", "staged_path": "figures/rendered/historical/v1/fig_forgetting_matrix.png", "staged_sha256": "58be85cb2527aa44c3e8c2579831138104bc70fa7e33794e1855df76988e9d6d", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 98630, "source_path": "experiments/dflash/figs/fig_fusion_combined.png", "source_sha256": "73c7ceff7bf8f01c0bc3e861c689532619dc2f358f51b37a0d9cedd8b07779bc", "staged_path": "figures/rendered/historical/v1/fig_fusion_combined.png", "staged_sha256": "73c7ceff7bf8f01c0bc3e861c689532619dc2f358f51b37a0d9cedd8b07779bc", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 163009, "source_path": "experiments/dflash/figs/fig_gen800k_loss_al.png", "source_sha256": "cc68719d712a0bc37321c46ec7d9b6368c08fc3f4523ef1d1c3d7771e5a99a37", "staged_path": "figures/rendered/historical/v1/fig_gen800k_loss_al.png", "staged_sha256": "cc68719d712a0bc37321c46ec7d9b6368c08fc3f4523ef1d1c3d7771e5a99a37", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 99807, "source_path": "experiments/dflash/figs/fig_merged_vs_specialist.png", "source_sha256": "ca381bfb16abfde13f5e88c917b27698155fa073e99cba70ad50235c6565a155", "staged_path": "figures/rendered/historical/v1/fig_merged_vs_specialist.png", "staged_sha256": "ca381bfb16abfde13f5e88c917b27698155fa073e99cba70ad50235c6565a155", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 109084, "source_path": "experiments/dflash/figs/fig_q3_4b_traj.png", "source_sha256": "16825387c4de0251447e50ba54aa99585cf8fc2790b1d0f4c8438035c40e23f0", "staged_path": "figures/rendered/historical/v1/fig_q3_4b_traj.png", "staged_sha256": "16825387c4de0251447e50ba54aa99585cf8fc2790b1d0f4c8438035c40e23f0", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 107592, "source_path": "experiments/dflash/figs/fig_serving.png", "source_sha256": "77ae4d2ff2f302212f1252b8bd1fbdc3c0d244dfd2a1be06d093a964575eff8e", "staged_path": "figures/rendered/historical/v1/fig_serving.png", "staged_sha256": "77ae4d2ff2f302212f1252b8bd1fbdc3c0d244dfd2a1be06d093a964575eff8e", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 73631, "source_path": "experiments/dflash/figs/fig_smalldata_vs_gen.png", "source_sha256": "1b8928ad66e0c86b58bc8d9088af11647e78f4f42180e95a0a5bd96017f623e3", "staged_path": "figures/rendered/historical/v1/fig_smalldata_vs_gen.png", "staged_sha256": "1b8928ad66e0c86b58bc8d9088af11647e78f4f42180e95a0a5bd96017f623e3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 57144, "source_path": "experiments/dflash/figs/fig_specialist_overall.png", "source_sha256": "0e3e6ae946db8f2922832f3b0b09673877c0cf68a55f9027fdc93d60399b9446", "staged_path": "figures/rendered/historical/v1/fig_specialist_overall.png", "staged_sha256": "0e3e6ae946db8f2922832f3b0b09673877c0cf68a55f9027fdc93d60399b9446", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 117399, "source_path": "experiments/dflash/figs/fig_warm_saturation.png", "source_sha256": "861c7f202252b6798391fb5bc98bc88d7379e38b3c7d77d2e4fbf0f74f144085", "staged_path": "figures/rendered/historical/v1/fig_warm_saturation.png", "staged_sha256": "861c7f202252b6798391fb5bc98bc88d7379e38b3c7d77d2e4fbf0f74f144085", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 34837, "source_path": "experiments/dflash/figs/gen_per_domain_AL.png", "source_sha256": "eb2084ef355799b98ab08000994e54aa8966196cbb3485235a942ed8966ac32c", "staged_path": "figures/rendered/historical/v1/gen_per_domain_AL.png", "staged_sha256": "eb2084ef355799b98ab08000994e54aa8966196cbb3485235a942ed8966ac32c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 136072, "source_path": "experiments/dflash/figs/gen_vs_merge_attn.png", "source_sha256": "2f484b361bed4165c4aac24a52ad66b7a3f9ea7948d7f617c764784085ac39b5", "staged_path": "figures/rendered/historical/v1/gen_vs_merge_attn.png", "staged_sha256": "2f484b361bed4165c4aac24a52ad66b7a3f9ea7948d7f617c764784085ac39b5", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 172019, "source_path": "experiments/dflash/figs/joint_vs_pipeline_code.png", "source_sha256": "221d21d5085f280ace83fe2532a3ae52858a6787af5f28addfd480095114984b", "staged_path": "figures/rendered/historical/v1/joint_vs_pipeline_code.png", "staged_sha256": "221d21d5085f280ace83fe2532a3ae52858a6787af5f28addfd480095114984b", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 222223, "source_path": "experiments/dflash/figs/overfit_attention_ladder.png", "source_sha256": "7bb2f21888c8c79e731add5c61021aff4be2c58a9dfc499601d902700867802c", "staged_path": "figures/rendered/historical/v1/overfit_attention_ladder.png", "staged_sha256": "7bb2f21888c8c79e731add5c61021aff4be2c58a9dfc499601d902700867802c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 181649, "source_path": "experiments/dflash/figs/overfit_loss_al.png", "source_sha256": "d8b24897fdce78a542aa0f9387e13299ffe3a34c679b607470ac984bddbbf10c", "staged_path": "figures/rendered/historical/v1/overfit_loss_al.png", "staged_sha256": "d8b24897fdce78a542aa0f9387e13299ffe3a34c679b607470ac984bddbbf10c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 96903, "source_path": "experiments/dflash/figs/reasonmix_5arm_per_domain.png", "source_sha256": "77b4ad8a26a01b67b052d7374de6b11350d89ba6eace1e8d135af159ffca5c7f", "staged_path": "figures/rendered/historical/v1/reasonmix_5arm_per_domain.png", "staged_sha256": "77b4ad8a26a01b67b052d7374de6b11350d89ba6eace1e8d135af159ffca5c7f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure", "license": "mit", "size_bytes": 359780, "source_path": "experiments/dflash/figs/recipe_budget.png", "source_sha256": "7c314a567c8cadfcfb602bc0ad10c328b6dca571213b88ef656bf78404350af4", "staged_path": "figures/rendered/historical/v1/recipe_budget.png", "staged_sha256": "7c314a567c8cadfcfb602bc0ad10c328b6dca571213b88ef656bf78404350af4", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure-or-source", "license": "mit", "size_bytes": 92555, "source_path": "paper/submission/figures/fig_architecture_v8.drawio", "source_sha256": "0415d8aa007c14b576bb25e56837b95235cf9e68afe828a775be58f265b46cb9", "staged_path": "figures/rendered/paper-current-candidate/v1/fig_architecture_v8.drawio", "staged_sha256": "0415d8aa007c14b576bb25e56837b95235cf9e68afe828a775be58f265b46cb9", "status": "paper-current-candidate-at-freeze", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure-or-source", "license": "mit", "size_bytes": 1543988, "source_path": "paper/submission/figures/fig_architecture_v8.png", "source_sha256": "3961c8a6ccc80992197b8746453cffd1d6f87ed158ce6103a72b4db361996f66", "staged_path": "figures/rendered/paper-current-candidate/v1/fig_architecture_v8.png", "staged_sha256": "3961c8a6ccc80992197b8746453cffd1d6f87ed158ce6103a72b4db361996f66", "status": "paper-current-candidate-at-freeze", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure-or-source", "license": "mit", "size_bytes": 232748, "source_path": "paper/submission/figures/fig_architecture_v8.svg", "source_sha256": "0a8c69fabf7ca2a57a13b4dd77edf2b94bfa1fd31fd9f28a259f7b014ec07048", "staged_path": "figures/rendered/paper-current-candidate/v1/fig_architecture_v8.svg", "staged_sha256": "0a8c69fabf7ca2a57a13b4dd77edf2b94bfa1fd31fd9f28a259f7b014ec07048", "status": "paper-current-candidate-at-freeze", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure-or-source", "license": "mit", "size_bytes": 450639, "source_path": "paper/submission/figures/fig_main_results.png", "source_sha256": "dc224065a687b7631cec8da84803ec6b8b1c5c34b737dfaa4e7077c9eb04318d", "staged_path": "figures/rendered/paper-current-candidate/v1/fig_main_results.png", "staged_sha256": "dc224065a687b7631cec8da84803ec6b8b1c5c34b737dfaa4e7077c9eb04318d", "status": "paper-current-candidate-at-freeze", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure-or-source", "license": "mit", "size_bytes": 23701, "source_path": "paper/submission/figures/fig_qwen3_4b_matched.pdf", "source_sha256": "c843a6bd2b0a20aad7d6b2816277b60af03b44bfd4a1a8a75bc7ac1c74bac746", "staged_path": "figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.pdf", "staged_sha256": "c843a6bd2b0a20aad7d6b2816277b60af03b44bfd4a1a8a75bc7ac1c74bac746", "status": "paper-current-candidate-at-freeze", "transformation": "direct-copy"} +{"artifact_type": "rendered-figure-or-source", "license": "mit", "size_bytes": 230257, "source_path": "paper/submission/figures/fig_qwen3_4b_matched.png", "source_sha256": "1243e3db541c0be5dbdacdabf6046ad065911ec2080bacd7d6e69bc9100689db", "staged_path": "figures/rendered/paper-current-candidate/v1/fig_qwen3_4b_matched.png", "staged_sha256": "1243e3db541c0be5dbdacdabf6046ad065911ec2080bacd7d6e69bc9100689db", "status": "paper-current-candidate-at-freeze", "transformation": "direct-copy"} +{"artifact_type": "license", "license": "mit", "size_bytes": 1068, "source_path": "LICENSE", "source_sha256": "da25818d68a3da676fed16a019beadd6b12f5e4d59ade85840c23a6070240323", "staged_path": "licenses/extended-evidence-v1/LICENSE", "staged_sha256": "da25818d68a3da676fed16a019beadd6b12f5e4d59ade85840c23a6070240323", "status": "current", "transformation": "direct-copy"} +{"artifact_type": "release-manifest", "license": "mit", "size_bytes": 1165, "source_path": null, "source_sha256": null, "staged_path": "manifests/extended-evidence-v1/release-summary.json", "staged_sha256": "ea5920812d91ae4a6e6cc503936d1c87fb833b622ef2b5bb9fb24037a4e5195e", "status": "current", "transformation": "generated-from-frozen-records"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 32852, "source_path": "experiments/dflash/scripts/build_mos_architecture_drawio.py", "source_sha256": "fe3d7a8074dc76533eefa49e627d779650c811fb0068698546330d811b8cb3c5", "staged_path": "recipes/plotting/v1/build_mos_architecture_drawio.py", "staged_sha256": "fe3d7a8074dc76533eefa49e627d779650c811fb0068698546330d811b8cb3c5", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 2500, "source_path": "experiments/dflash/scripts/fig_5arm_per_domain.py", "source_sha256": "00973527df3dc197e5946b12681d86a5a469e04b43650865e5b9820513693e9f", "staged_path": "recipes/plotting/v1/fig_5arm_per_domain.py", "staged_sha256": "c8e6297def7e8dfc361a437c798c8b8586f6f0ee99772cfa915ecc325ea9ba01", "status": "reusable-code", "transformation": "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 1387, "source_path": "experiments/dflash/scripts/fig_code250k_vs_gen.py", "source_sha256": "19569991010d153e28b83437fc1b075748734c233cb2ece176d23de424e0cda6", "staged_path": "recipes/plotting/v1/fig_code250k_vs_gen.py", "staged_sha256": "c6ec2787808a900478e9a92fb7f3092e04835603882d39a3ad0d24b2d486e193", "status": "reusable-code", "transformation": "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 3439, "source_path": "experiments/dflash/scripts/fig_exp1_forgetting.py", "source_sha256": "687845a1e7bb9adf56a80d7467ee219a7044884a2163dfa6c18af5a6153b9b29", "staged_path": "recipes/plotting/v1/fig_exp1_forgetting.py", "staged_sha256": "f49d9a5dfc3f7512af3cc86f0729774344357fa7de748e9917e90a5a4bf16d84", "status": "reusable-code", "transformation": "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 1830, "source_path": "experiments/dflash/scripts/fig_exp1_three_recipes.py", "source_sha256": "764423e19ac26e5e0fdc57e255b94dd20d7389d3f91eed439a5aaab38280690d", "staged_path": "recipes/plotting/v1/fig_exp1_three_recipes.py", "staged_sha256": "9fd98d1f1bb1a6e449bf503569fd8fbd82d89f7366a811f0fd9be4e3b9ac7e95", "status": "reusable-code", "transformation": "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 2779, "source_path": "experiments/dflash/scripts/fig_exp4_inference.py", "source_sha256": "d12deefede0eaf5f3b046167ebbbe51bef2db6a919524a9463eba4db7c16937c", "staged_path": "recipes/plotting/v1/fig_exp4_inference.py", "staged_sha256": "ebe9bc1be8f0de7bdd83bcc74e71314fd9ca7758802bfa660128a265c5e4f9f1", "status": "reusable-code", "transformation": "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 2453, "source_path": "experiments/dflash/scripts/fig_exp5_method_vs_data.py", "source_sha256": "6b6056f34fca602e695b1e430a67d840b64023d1e5c1226274a3358dc799eaac", "staged_path": "recipes/plotting/v1/fig_exp5_method_vs_data.py", "staged_sha256": "44f959ce1232a19ea3dcee93bb41df5268171e47cf2d66ed4b5e7d84312497a5", "status": "reusable-code", "transformation": "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 2316, "source_path": "experiments/dflash/scripts/fig_fusion_combined.py", "source_sha256": "ca2febdec76a27d61ae1f38a68834150b03639173967ef832470d011aeff8539", "staged_path": "recipes/plotting/v1/fig_fusion_combined.py", "staged_sha256": "39542922ee86c359fc3c6e0153daba478ada98e8d10da5e40b8b9994a440a87f", "status": "reusable-code", "transformation": "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 6868, "source_path": "experiments/dflash/scripts/fig_matrix_and_arms.py", "source_sha256": "1e59dd91ae95f5b348aff7fb7d99a44bd8808e3b2ebb6e0db3959656235d6510", "staged_path": "recipes/plotting/v1/fig_matrix_and_arms.py", "staged_sha256": "6967a36b0a4ce87c5650409a561bc90a186eb25633501593c87e6da9460c2bd2", "status": "reusable-code", "transformation": "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 2270, "source_path": "experiments/dflash/scripts/fig_merged_vs_specialist.py", "source_sha256": "aff7b087d456145e5e12f7d20bee533e4eeb8c1281ca073ce00db064e19ce2f0", "staged_path": "recipes/plotting/v1/fig_merged_vs_specialist.py", "staged_sha256": "6eb66c23518bd7e2b8edc74f23ddb53635cac18ae839c4862ec9bfbc1ca8bdc5", "status": "reusable-code", "transformation": "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 3405, "source_path": "experiments/dflash/scripts/fig_serving_specialist.py", "source_sha256": "6012b78226d9a95c89faf03d26975e098c90a86ff9f76a04e0f2149f3f29fc2e", "staged_path": "recipes/plotting/v1/fig_serving_specialist.py", "staged_sha256": "a692b7e495c6d3ceaf46857436a115ffc299375936e37f820ddda02468a724b7", "status": "reusable-code", "transformation": "sanitized-copy; replaced 2 hard-coded output path(s) with relative output path(s)"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 11237, "source_path": "experiments/dflash/scripts/plot_main_results.py", "source_sha256": "d28b0d29bda5f5ee10c4ea22fa24e7b1b07e5ba2d86c57f74f58285b06e9b58e", "staged_path": "recipes/plotting/v1/plot_main_results.py", "staged_sha256": "d28b0d29bda5f5ee10c4ea22fa24e7b1b07e5ba2d86c57f74f58285b06e9b58e", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 30875, "source_path": "experiments/dflash/scripts/plot_mos_5x5_gains.py", "source_sha256": "663f813697a0026b093a0a6ef68d3b99be3a7c159eadbe2c5b3522e645d92eec", "staged_path": "recipes/plotting/v1/plot_mos_5x5_gains.py", "staged_sha256": "663f813697a0026b093a0a6ef68d3b99be3a7c159eadbe2c5b3522e645d92eec", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 5846, "source_path": "experiments/dflash/scripts/plot_qwen3_4b_matched.py", "source_sha256": "6fdf9f53c384d1ec47ed182dcf2d44987ff30bb0255bef0061fa7c723b785c0e", "staged_path": "recipes/plotting/v1/plot_qwen3_4b_matched.py", "staged_sha256": "6fdf9f53c384d1ec47ed182dcf2d44987ff30bb0255bef0061fa7c723b785c0e", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "plotting-recipe", "license": "mit", "size_bytes": 6951, "source_path": "experiments/dflash/scripts/plot_recipe_budget.py", "source_sha256": "a7beb62c3b84008553ff9051ecc76bbd859667bf28898b182317363c8128e404", "staged_path": "recipes/plotting/v1/plot_recipe_budget.py", "staged_sha256": "a7beb62c3b84008553ff9051ecc76bbd859667bf28898b182317363c8128e404", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "environment-recipe", "license": "mit", "size_bytes": 35, "source_path": null, "source_sha256": null, "staged_path": "recipes/requirements-v1.txt", "staged_sha256": "b7ebb2c81dacb6ef06dcb87fb879ea663f2327b7417606520bb30cb299ef006a", "status": "minimum-unpinned-dependencies", "transformation": "generated-from-selected-script-imports"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/a2_code/al_curve.csv", "source_sha256": "dae17f81b9fa4b5dba14545e732f6731e359e80e3845f937c09fe81ef8a6e068", "staged_path": "results/historical-al-curves/v1/a2_code/al_curve.csv", "staged_sha256": "dae17f81b9fa4b5dba14545e732f6731e359e80e3845f937c09fe81ef8a6e068", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/a2_creative_writing/al_curve.csv", "source_sha256": "7d56cbca48fdec9cc37fc26e701ca5da6841884f8a49e51c929090c49cb5fe6b", "staged_path": "results/historical-al-curves/v1/a2_creative_writing/al_curve.csv", "staged_sha256": "7d56cbca48fdec9cc37fc26e701ca5da6841884f8a49e51c929090c49cb5fe6b", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/a2_factual_qa/al_curve.csv", "source_sha256": "5da7a3f0e5d955881ce5903f4a7b7567212f89400f130a574fdc0cb783169551", "staged_path": "results/historical-al-curves/v1/a2_factual_qa/al_curve.csv", "staged_sha256": "5da7a3f0e5d955881ce5903f4a7b7567212f89400f130a574fdc0cb783169551", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/a2_general/al_curve.csv", "source_sha256": "aee9cf5c10f54cadba4b30b513c18b2651f1fbc6b10d25da719eaea74f471682", "staged_path": "results/historical-al-curves/v1/a2_general/al_curve.csv", "staged_sha256": "aee9cf5c10f54cadba4b30b513c18b2651f1fbc6b10d25da719eaea74f471682", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/a2_math/al_curve.csv", "source_sha256": "4ce95c7937d1362f0593ad93531d65620a484e138237ce793ea1b3ef03099c5b", "staged_path": "results/historical-al-curves/v1/a2_math/al_curve.csv", "staged_sha256": "4ce95c7937d1362f0593ad93531d65620a484e138237ce793ea1b3ef03099c5b", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/a3_code/al_curve.csv", "source_sha256": "e88672fadfaaaeb1088504c919208f8b125f1aecb7a14942fc5888d572714857", "staged_path": "results/historical-al-curves/v1/a3_code/al_curve.csv", "staged_sha256": "e88672fadfaaaeb1088504c919208f8b125f1aecb7a14942fc5888d572714857", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/a3_creative_writing/al_curve.csv", "source_sha256": "b5a492456138dc23d09264c26ca671082eb3fb36765ec5398aa31f56d77d2cc1", "staged_path": "results/historical-al-curves/v1/a3_creative_writing/al_curve.csv", "staged_sha256": "b5a492456138dc23d09264c26ca671082eb3fb36765ec5398aa31f56d77d2cc1", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/a3_factual_qa/al_curve.csv", "source_sha256": "b7b8776a83582db069ecd349784aa52d40dcd3bbd6c100c13780a873590c4b98", "staged_path": "results/historical-al-curves/v1/a3_factual_qa/al_curve.csv", "staged_sha256": "b7b8776a83582db069ecd349784aa52d40dcd3bbd6c100c13780a873590c4b98", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/a3_general/al_curve.csv", "source_sha256": "4073a14f31f2d26f0c1974f3e571a45c24209bf2562b2c7725d207d134047a9e", "staged_path": "results/historical-al-curves/v1/a3_general/al_curve.csv", "staged_sha256": "4073a14f31f2d26f0c1974f3e571a45c24209bf2562b2c7725d207d134047a9e", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/a3_math/al_curve.csv", "source_sha256": "07ba1e5ce58ff013aab9a96738301295f32731facc3ec5b7678802d5ef8c996f", "staged_path": "results/historical-al-curves/v1/a3_math/al_curve.csv", "staged_sha256": "07ba1e5ce58ff013aab9a96738301295f32731facc3ec5b7678802d5ef8c996f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/a4_code/al_curve.csv", "source_sha256": "cd427bc83af9d2f892924bd1993ead327acab61c06279e5c627abe2e18cc6809", "staged_path": "results/historical-al-curves/v1/a4_code/al_curve.csv", "staged_sha256": "cd427bc83af9d2f892924bd1993ead327acab61c06279e5c627abe2e18cc6809", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/a4_creative_writing/al_curve.csv", "source_sha256": "c1867fc99b121cfd613ce2a9d40f8f12a13491e146f85d3175d7bf8c3591c02f", "staged_path": "results/historical-al-curves/v1/a4_creative_writing/al_curve.csv", "staged_sha256": "c1867fc99b121cfd613ce2a9d40f8f12a13491e146f85d3175d7bf8c3591c02f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/a4_factual_qa/al_curve.csv", "source_sha256": "cf142156c17e74d5ac67967b93995dbaf47ee34d4ab28cf65d0044846cd52785", "staged_path": "results/historical-al-curves/v1/a4_factual_qa/al_curve.csv", "staged_sha256": "cf142156c17e74d5ac67967b93995dbaf47ee34d4ab28cf65d0044846cd52785", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/a4_general/al_curve.csv", "source_sha256": "81d3dcaacdd7405966d5eb5715dd04c434e714abf3a9620ae492430065c8dbc6", "staged_path": "results/historical-al-curves/v1/a4_general/al_curve.csv", "staged_sha256": "81d3dcaacdd7405966d5eb5715dd04c434e714abf3a9620ae492430065c8dbc6", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/a4_math/al_curve.csv", "source_sha256": "8e717efab00e869e764262597696297a9bfd5cd7839e0d9ddd2ba623acb0aa05", "staged_path": "results/historical-al-curves/v1/a4_math/al_curve.csv", "staged_sha256": "8e717efab00e869e764262597696297a9bfd5cd7839e0d9ddd2ba623acb0aa05", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/a5_code/al_curve.csv", "source_sha256": "c105111e04a724b5b21af90fd5e1ecd2b10820db32c9fb041e3cfe153e862f6f", "staged_path": "results/historical-al-curves/v1/a5_code/al_curve.csv", "staged_sha256": "c105111e04a724b5b21af90fd5e1ecd2b10820db32c9fb041e3cfe153e862f6f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/a5_creative_writing/al_curve.csv", "source_sha256": "454d8a6f5a03494e4d8aace991965482a858e04bbcdd4c16cb07773ba3895afa", "staged_path": "results/historical-al-curves/v1/a5_creative_writing/al_curve.csv", "staged_sha256": "454d8a6f5a03494e4d8aace991965482a858e04bbcdd4c16cb07773ba3895afa", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 85, "source_path": "experiments/dflash/eval_csv/a5_factual_qa/al_curve.csv", "source_sha256": "43df7e88a7ddd7bbbc9471c65682b4584af117b0d1f02dc0ade07bb29395c9cf", "staged_path": "results/historical-al-curves/v1/a5_factual_qa/al_curve.csv", "staged_sha256": "43df7e88a7ddd7bbbc9471c65682b4584af117b0d1f02dc0ade07bb29395c9cf", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/a5_general/al_curve.csv", "source_sha256": "e64d977591372dca44bb442ced4a6f5c1f39464a84471bf1ebc21abe5e29bdd3", "staged_path": "results/historical-al-curves/v1/a5_general/al_curve.csv", "staged_sha256": "e64d977591372dca44bb442ced4a6f5c1f39464a84471bf1ebc21abe5e29bdd3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/a5_math/al_curve.csv", "source_sha256": "b85e52e687ba97220b8cad613cc209578df53767e40ab55fd9cf2fdf5808f3b1", "staged_path": "results/historical-al-curves/v1/a5_math/al_curve.csv", "staged_sha256": "b85e52e687ba97220b8cad613cc209578df53767e40ab55fd9cf2fdf5808f3b1", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/a6_code/al_curve.csv", "source_sha256": "0784056349eac36c44dc1da700b5e051dd60ebfb832254fadf0e30049bdb6644", "staged_path": "results/historical-al-curves/v1/a6_code/al_curve.csv", "staged_sha256": "0784056349eac36c44dc1da700b5e051dd60ebfb832254fadf0e30049bdb6644", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/a6_creative_writing/al_curve.csv", "source_sha256": "6f0c6689cc637a4046f1cd36d32c97f5df4d6931d4bbd0df7f373929a7384258", "staged_path": "results/historical-al-curves/v1/a6_creative_writing/al_curve.csv", "staged_sha256": "6f0c6689cc637a4046f1cd36d32c97f5df4d6931d4bbd0df7f373929a7384258", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/a6_factual_qa/al_curve.csv", "source_sha256": "8ab6645dd36b4c7ed6e150591aa796d07fca2f453083a4ea7690b19d14f0522c", "staged_path": "results/historical-al-curves/v1/a6_factual_qa/al_curve.csv", "staged_sha256": "8ab6645dd36b4c7ed6e150591aa796d07fca2f453083a4ea7690b19d14f0522c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/a6_general/al_curve.csv", "source_sha256": "dcc0ad60fb88da0a1c2b124b7e6daa94fa4bea76de8ac393eac3edefd900de8d", "staged_path": "results/historical-al-curves/v1/a6_general/al_curve.csv", "staged_sha256": "dcc0ad60fb88da0a1c2b124b7e6daa94fa4bea76de8ac393eac3edefd900de8d", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/a6_math/al_curve.csv", "source_sha256": "743758c18a447dc760c512d1525a6adea57a0f8086d0d0b7b8acf0a88a9b16b7", "staged_path": "results/historical-al-curves/v1/a6_math/al_curve.csv", "staged_sha256": "743758c18a447dc760c512d1525a6adea57a0f8086d0d0b7b8acf0a88a9b16b7", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/a7_code/al_curve.csv", "source_sha256": "c3c3784e3a7046135fe0772b9e9acedbfe45473bf125a7c69766e08118a90b9f", "staged_path": "results/historical-al-curves/v1/a7_code/al_curve.csv", "staged_sha256": "c3c3784e3a7046135fe0772b9e9acedbfe45473bf125a7c69766e08118a90b9f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 97, "source_path": "experiments/dflash/eval_csv/a7_creative_writing/al_curve.csv", "source_sha256": "0eab7d6c41d11ccd0653115a7077d4b9e22943bbd9512867b71eaa2e0df01327", "staged_path": "results/historical-al-curves/v1/a7_creative_writing/al_curve.csv", "staged_sha256": "0eab7d6c41d11ccd0653115a7077d4b9e22943bbd9512867b71eaa2e0df01327", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/a7_factual_qa/al_curve.csv", "source_sha256": "d86f5a9a467df89b88deb48595e53915114626e692200afe2b1cceb68951f7ad", "staged_path": "results/historical-al-curves/v1/a7_factual_qa/al_curve.csv", "staged_sha256": "d86f5a9a467df89b88deb48595e53915114626e692200afe2b1cceb68951f7ad", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/a7_general/al_curve.csv", "source_sha256": "c66d5d71f44d502be2f4319b57d99d7ab498e470e8b29808fc7943c5cde07423", "staged_path": "results/historical-al-curves/v1/a7_general/al_curve.csv", "staged_sha256": "c66d5d71f44d502be2f4319b57d99d7ab498e470e8b29808fc7943c5cde07423", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/a7_math/al_curve.csv", "source_sha256": "07fa6f01a556aff84978b938d7210e2c32cbcea0ff28b672895169ef2096b9fa", "staged_path": "results/historical-al-curves/v1/a7_math/al_curve.csv", "staged_sha256": "07fa6f01a556aff84978b938d7210e2c32cbcea0ff28b672895169ef2096b9fa", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/aw0_code/al_curve.csv", "source_sha256": "63bc1738955954356859cdca199f48a3948003f6c6ceb523e7658dc6297c4fe5", "staged_path": "results/historical-al-curves/v1/aw0_code/al_curve.csv", "staged_sha256": "63bc1738955954356859cdca199f48a3948003f6c6ceb523e7658dc6297c4fe5", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 100, "source_path": "experiments/dflash/eval_csv/aw1_code/al_curve.csv", "source_sha256": "58a7c10f203fdfca83c4445aab6b8db5bc93a5ebf6c5ee39a30adea955238048", "staged_path": "results/historical-al-curves/v1/aw1_code/al_curve.csv", "staged_sha256": "58a7c10f203fdfca83c4445aab6b8db5bc93a5ebf6c5ee39a30adea955238048", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 101, "source_path": "experiments/dflash/eval_csv/aw2_code/al_curve.csv", "source_sha256": "766d1a00823d6db3f2cd6a3d96e11b935d31120f95f780b6d1f8b41a53a95f82", "staged_path": "results/historical-al-curves/v1/aw2_code/al_curve.csv", "staged_sha256": "766d1a00823d6db3f2cd6a3d96e11b935d31120f95f780b6d1f8b41a53a95f82", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 101, "source_path": "experiments/dflash/eval_csv/aw3_code/al_curve.csv", "source_sha256": "b3b6db437eb990291582e1fe8180fb77a5fc44394ae3a4bd595b58a1fa95116e", "staged_path": "results/historical-al-curves/v1/aw3_code/al_curve.csv", "staged_sha256": "b3b6db437eb990291582e1fe8180fb77a5fc44394ae3a4bd595b58a1fa95116e", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 101, "source_path": "experiments/dflash/eval_csv/aw4_code/al_curve.csv", "source_sha256": "71a499635ea64fb0e45ea136aafe69f2466b508e0b66e6b8158a681ab0fb6ffb", "staged_path": "results/historical-al-curves/v1/aw4_code/al_curve.csv", "staged_sha256": "71a499635ea64fb0e45ea136aafe69f2466b508e0b66e6b8158a681ab0fb6ffb", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 101, "source_path": "experiments/dflash/eval_csv/aw5_code/al_curve.csv", "source_sha256": "e135392f04e9db93600e871c8a82b556b7f380c7622db82489bf158358fb0a29", "staged_path": "results/historical-al-curves/v1/aw5_code/al_curve.csv", "staged_sha256": "e135392f04e9db93600e871c8a82b556b7f380c7622db82489bf158358fb0a29", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 100, "source_path": "experiments/dflash/eval_csv/aw6_code/al_curve.csv", "source_sha256": "23bdd93d5bfa5c24aed9089fb1f1903927a2de4bf26629ee42194d31994fe2c3", "staged_path": "results/historical-al-curves/v1/aw6_code/al_curve.csv", "staged_sha256": "23bdd93d5bfa5c24aed9089fb1f1903927a2de4bf26629ee42194d31994fe2c3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 101, "source_path": "experiments/dflash/eval_csv/aw7_code/al_curve.csv", "source_sha256": "793994b7cbf0695dcc184bf095697988aef78be0ad6c8fabb615c2f16200d652", "staged_path": "results/historical-al-curves/v1/aw7_code/al_curve.csv", "staged_sha256": "793994b7cbf0695dcc184bf095697988aef78be0ad6c8fabb615c2f16200d652", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/b0_code/al_curve.csv", "source_sha256": "ed2bab825dda5d70ddb6a3e16a0a86633178f1c8e3a2149031d56f11be4b05b0", "staged_path": "results/historical-al-curves/v1/b0_code/al_curve.csv", "staged_sha256": "ed2bab825dda5d70ddb6a3e16a0a86633178f1c8e3a2149031d56f11be4b05b0", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 97, "source_path": "experiments/dflash/eval_csv/b0_creative_writing/al_curve.csv", "source_sha256": "f56a485034b29f7e1c65c8021efe18f5980d732c4e1556d6550e93a52d55f616", "staged_path": "results/historical-al-curves/v1/b0_creative_writing/al_curve.csv", "staged_sha256": "f56a485034b29f7e1c65c8021efe18f5980d732c4e1556d6550e93a52d55f616", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/b0_factual_qa/al_curve.csv", "source_sha256": "f33c9d7fa3985e39ac50ff1d8dad02d72e7276a08c8d43492627be23ed67c579", "staged_path": "results/historical-al-curves/v1/b0_factual_qa/al_curve.csv", "staged_sha256": "f33c9d7fa3985e39ac50ff1d8dad02d72e7276a08c8d43492627be23ed67c579", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/b0_general/al_curve.csv", "source_sha256": "51ea0ab13244e62257a02239ca42aade659a535afff79faafc4f37c4d6ca5316", "staged_path": "results/historical-al-curves/v1/b0_general/al_curve.csv", "staged_sha256": "51ea0ab13244e62257a02239ca42aade659a535afff79faafc4f37c4d6ca5316", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 73, "source_path": "experiments/dflash/eval_csv/b0_math/al_curve.csv", "source_sha256": "405dd523d42728d330a5a5de80f12957a15f954c41f06474c0f898f841c7b4b9", "staged_path": "results/historical-al-curves/v1/b0_math/al_curve.csv", "staged_sha256": "405dd523d42728d330a5a5de80f12957a15f954c41f06474c0f898f841c7b4b9", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/b1_code/al_curve.csv", "source_sha256": "97932cc47a49befbf27f7957d827a37c3b93d691d8b5692562c9fe52d964d941", "staged_path": "results/historical-al-curves/v1/b1_code/al_curve.csv", "staged_sha256": "97932cc47a49befbf27f7957d827a37c3b93d691d8b5692562c9fe52d964d941", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/b1_creative_writing/al_curve.csv", "source_sha256": "f8185aae5e4724c5a86c6adfeb34c16e7767f929d3d633c0d60cab0f8127128b", "staged_path": "results/historical-al-curves/v1/b1_creative_writing/al_curve.csv", "staged_sha256": "f8185aae5e4724c5a86c6adfeb34c16e7767f929d3d633c0d60cab0f8127128b", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/b1_factual_qa/al_curve.csv", "source_sha256": "c2031c105435638bb133232dc1bd0c8b23454a368fa6bddc457426981dfbde56", "staged_path": "results/historical-al-curves/v1/b1_factual_qa/al_curve.csv", "staged_sha256": "c2031c105435638bb133232dc1bd0c8b23454a368fa6bddc457426981dfbde56", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/b1_general/al_curve.csv", "source_sha256": "0967339a7d6711ef84490506221730589487c1d21586cd1e131bb4b5e0540a11", "staged_path": "results/historical-al-curves/v1/b1_general/al_curve.csv", "staged_sha256": "0967339a7d6711ef84490506221730589487c1d21586cd1e131bb4b5e0540a11", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/b1_math/al_curve.csv", "source_sha256": "1fb5f0f30ded6faad7a1f8a02050853cb500df707853dfbfe4239016e625a6bf", "staged_path": "results/historical-al-curves/v1/b1_math/al_curve.csv", "staged_sha256": "1fb5f0f30ded6faad7a1f8a02050853cb500df707853dfbfe4239016e625a6bf", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/b2_code/al_curve.csv", "source_sha256": "784723990cdf7fb9cbae73598ade59da5c5e494666187d39ccab98e2a13e843c", "staged_path": "results/historical-al-curves/v1/b2_code/al_curve.csv", "staged_sha256": "784723990cdf7fb9cbae73598ade59da5c5e494666187d39ccab98e2a13e843c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/b2_creative_writing/al_curve.csv", "source_sha256": "2d7c6e582ced15922331dbd31ca93ac17476daa07af395bba9f2c4e90793ada8", "staged_path": "results/historical-al-curves/v1/b2_creative_writing/al_curve.csv", "staged_sha256": "2d7c6e582ced15922331dbd31ca93ac17476daa07af395bba9f2c4e90793ada8", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/b2_factual_qa/al_curve.csv", "source_sha256": "8a7fc49463e125c3572f43300ae6f51229d3b7a9460f8a4cd7415f4029155e0d", "staged_path": "results/historical-al-curves/v1/b2_factual_qa/al_curve.csv", "staged_sha256": "8a7fc49463e125c3572f43300ae6f51229d3b7a9460f8a4cd7415f4029155e0d", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/b2_general/al_curve.csv", "source_sha256": "17a09873412bd568cc0e707da3502f5413a58fc10085a640d4e6433645ad0307", "staged_path": "results/historical-al-curves/v1/b2_general/al_curve.csv", "staged_sha256": "17a09873412bd568cc0e707da3502f5413a58fc10085a640d4e6433645ad0307", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/b2_math/al_curve.csv", "source_sha256": "8cfcdd3707b7474195e9c1d55361ebf1d7798eccc1bd5d2e77a5e354a05e0272", "staged_path": "results/historical-al-curves/v1/b2_math/al_curve.csv", "staged_sha256": "8cfcdd3707b7474195e9c1d55361ebf1d7798eccc1bd5d2e77a5e354a05e0272", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 387, "source_path": "experiments/dflash/eval_csv/b2_v3_cont_curve_8x5x1x6/al_curve.csv", "source_sha256": "e4f9520174ba9ba33873c3c667883034cb9112e0ed8fb1e9ec6f680142ded0c5", "staged_path": "results/historical-al-curves/v1/b2_v3_cont_curve_8x5x1x6/al_curve.csv", "staged_sha256": "e4f9520174ba9ba33873c3c667883034cb9112e0ed8fb1e9ec6f680142ded0c5", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 853, "source_path": "experiments/dflash/eval_csv/b2_v3_curve_8x5x1x6/al_curve.csv", "source_sha256": "c36f8911c0042f6f686cd63a43330bb2dcb007709aa8b15807eb8a32dd259341", "staged_path": "results/historical-al-curves/v1/b2_v3_curve_8x5x1x6/al_curve.csv", "staged_sha256": "c36f8911c0042f6f686cd63a43330bb2dcb007709aa8b15807eb8a32dd259341", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/b3_code/al_curve.csv", "source_sha256": "0944a8604c0bd5a5218a9a0c6a4b3ed370657b300e8941d9a5265f5b21634b54", "staged_path": "results/historical-al-curves/v1/b3_code/al_curve.csv", "staged_sha256": "0944a8604c0bd5a5218a9a0c6a4b3ed370657b300e8941d9a5265f5b21634b54", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/b3_creative_writing/al_curve.csv", "source_sha256": "7f35dd8bac7bec1c7fd38eee49ef144cf8cd1ea6577c28ebf7b02008a0e018c4", "staged_path": "results/historical-al-curves/v1/b3_creative_writing/al_curve.csv", "staged_sha256": "7f35dd8bac7bec1c7fd38eee49ef144cf8cd1ea6577c28ebf7b02008a0e018c4", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/b3_factual_qa/al_curve.csv", "source_sha256": "393b855f4794d72ab7fd16d2dff264c975f3407f3b781831465fd79334e705f3", "staged_path": "results/historical-al-curves/v1/b3_factual_qa/al_curve.csv", "staged_sha256": "393b855f4794d72ab7fd16d2dff264c975f3407f3b781831465fd79334e705f3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/b3_general/al_curve.csv", "source_sha256": "76db113c70a264f7f3d96bb764b2befdcbd8ddd30593babbd13be47c7ccd36c2", "staged_path": "results/historical-al-curves/v1/b3_general/al_curve.csv", "staged_sha256": "76db113c70a264f7f3d96bb764b2befdcbd8ddd30593babbd13be47c7ccd36c2", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/b3_math/al_curve.csv", "source_sha256": "cb2c4c321389e1f978648ac9b634532606885924336ee9eff3e9ee3df0dd8806", "staged_path": "results/historical-al-curves/v1/b3_math/al_curve.csv", "staged_sha256": "cb2c4c321389e1f978648ac9b634532606885924336ee9eff3e9ee3df0dd8806", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/b4_code/al_curve.csv", "source_sha256": "20dcba4e0f4a97d64d795ab5e97e377d14f083ce3ca8e9febf6ab1def6bd05d8", "staged_path": "results/historical-al-curves/v1/b4_code/al_curve.csv", "staged_sha256": "20dcba4e0f4a97d64d795ab5e97e377d14f083ce3ca8e9febf6ab1def6bd05d8", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/b4_creative_writing/al_curve.csv", "source_sha256": "cfb4e10bdb3f8896f321403b653c01f8edc7be518376a7508ee1fe16abf5133b", "staged_path": "results/historical-al-curves/v1/b4_creative_writing/al_curve.csv", "staged_sha256": "cfb4e10bdb3f8896f321403b653c01f8edc7be518376a7508ee1fe16abf5133b", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/b4_factual_qa/al_curve.csv", "source_sha256": "6889a3c240b28f1708a8aa1844fbd9e016ede9cd06201d3f3444e852d6bd18a9", "staged_path": "results/historical-al-curves/v1/b4_factual_qa/al_curve.csv", "staged_sha256": "6889a3c240b28f1708a8aa1844fbd9e016ede9cd06201d3f3444e852d6bd18a9", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/b4_general/al_curve.csv", "source_sha256": "93e2192320c39883bb6e536275fbc5af9992c285856a98c51f745a6c2cd322f1", "staged_path": "results/historical-al-curves/v1/b4_general/al_curve.csv", "staged_sha256": "93e2192320c39883bb6e536275fbc5af9992c285856a98c51f745a6c2cd322f1", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/b4_math/al_curve.csv", "source_sha256": "897845f483d8fbbc7f9cf587698bc4b7a96e126f7716172bc5547f7138ba7bfd", "staged_path": "results/historical-al-curves/v1/b4_math/al_curve.csv", "staged_sha256": "897845f483d8fbbc7f9cf587698bc4b7a96e126f7716172bc5547f7138ba7bfd", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/b5_code/al_curve.csv", "source_sha256": "092f185463e80a3f3870785173b7047d62662919946293db45b2e432318717e5", "staged_path": "results/historical-al-curves/v1/b5_code/al_curve.csv", "staged_sha256": "092f185463e80a3f3870785173b7047d62662919946293db45b2e432318717e5", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/b5_creative_writing/al_curve.csv", "source_sha256": "0c193663d6bf65c6ffede829949cb80bc68801c4b52e7b3b254c410cb476e8d3", "staged_path": "results/historical-al-curves/v1/b5_creative_writing/al_curve.csv", "staged_sha256": "0c193663d6bf65c6ffede829949cb80bc68801c4b52e7b3b254c410cb476e8d3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/b5_factual_qa/al_curve.csv", "source_sha256": "bd5d632a8da9392f4ff22a73c1540e8990ba72407c09eb80e5f326abd31657d3", "staged_path": "results/historical-al-curves/v1/b5_factual_qa/al_curve.csv", "staged_sha256": "bd5d632a8da9392f4ff22a73c1540e8990ba72407c09eb80e5f326abd31657d3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/b5_general/al_curve.csv", "source_sha256": "7ab5552d57bee9349fa05b10f0ec20751db6d57745f192b0fff040df83c70db7", "staged_path": "results/historical-al-curves/v1/b5_general/al_curve.csv", "staged_sha256": "7ab5552d57bee9349fa05b10f0ec20751db6d57745f192b0fff040df83c70db7", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 73, "source_path": "experiments/dflash/eval_csv/b5_math/al_curve.csv", "source_sha256": "601ddb84962c2e9a12c49338d3688b534b1b6e8efc3ea1f7c47dfd8274ac3e85", "staged_path": "results/historical-al-curves/v1/b5_math/al_curve.csv", "staged_sha256": "601ddb84962c2e9a12c49338d3688b534b1b6e8efc3ea1f7c47dfd8274ac3e85", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/b6_code/al_curve.csv", "source_sha256": "d5858dc1897c4d64feadf087ea74740c3657b736e50b1561e442c7d7e66a0b3a", "staged_path": "results/historical-al-curves/v1/b6_code/al_curve.csv", "staged_sha256": "d5858dc1897c4d64feadf087ea74740c3657b736e50b1561e442c7d7e66a0b3a", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 96, "source_path": "experiments/dflash/eval_csv/b6_creative_writing/al_curve.csv", "source_sha256": "752306586324989229b99b0eb4dc42d52faca211e44337bd279f063ee27c31c8", "staged_path": "results/historical-al-curves/v1/b6_creative_writing/al_curve.csv", "staged_sha256": "752306586324989229b99b0eb4dc42d52faca211e44337bd279f063ee27c31c8", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/b6_factual_qa/al_curve.csv", "source_sha256": "97e815c6b08c33cd7c10581fed459dfe06d2a01328d1f433908f4ec0261b469f", "staged_path": "results/historical-al-curves/v1/b6_factual_qa/al_curve.csv", "staged_sha256": "97e815c6b08c33cd7c10581fed459dfe06d2a01328d1f433908f4ec0261b469f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/b6_general/al_curve.csv", "source_sha256": "9da6e16cbbe001a4826ad54233041c396ce4dce90e4074639f6a74ce9c4dcb62", "staged_path": "results/historical-al-curves/v1/b6_general/al_curve.csv", "staged_sha256": "9da6e16cbbe001a4826ad54233041c396ce4dce90e4074639f6a74ce9c4dcb62", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/b6_math/al_curve.csv", "source_sha256": "66c4a1fc30f00ec6a172c79df7bc9bfd5eb20300b4b45961387b215a3459f223", "staged_path": "results/historical-al-curves/v1/b6_math/al_curve.csv", "staged_sha256": "66c4a1fc30f00ec6a172c79df7bc9bfd5eb20300b4b45961387b215a3459f223", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/b7_code/al_curve.csv", "source_sha256": "9caaf63ef828990870bcfbc53c48d849cfd1b9a7db2f33b028a82f32e7246ccd", "staged_path": "results/historical-al-curves/v1/b7_code/al_curve.csv", "staged_sha256": "9caaf63ef828990870bcfbc53c48d849cfd1b9a7db2f33b028a82f32e7246ccd", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/b7_creative_writing/al_curve.csv", "source_sha256": "16228b3deb7f7e45f190ef229186df928daaa0a182b9a7f7b5a8fe0e0bf84748", "staged_path": "results/historical-al-curves/v1/b7_creative_writing/al_curve.csv", "staged_sha256": "16228b3deb7f7e45f190ef229186df928daaa0a182b9a7f7b5a8fe0e0bf84748", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/b7_factual_qa/al_curve.csv", "source_sha256": "6d380638c474c93a128897ea0d5e38507c8fc639cd5afdac85cfbf55e9be72ae", "staged_path": "results/historical-al-curves/v1/b7_factual_qa/al_curve.csv", "staged_sha256": "6d380638c474c93a128897ea0d5e38507c8fc639cd5afdac85cfbf55e9be72ae", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/b7_general/al_curve.csv", "source_sha256": "4c00e7142537032150da6be745d880fb00fc63e7722d3f016cc7801bc9180884", "staged_path": "results/historical-al-curves/v1/b7_general/al_curve.csv", "staged_sha256": "4c00e7142537032150da6be745d880fb00fc63e7722d3f016cc7801bc9180884", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/b7_math/al_curve.csv", "source_sha256": "ea0b445042ac886293657641a64bb039fd6e2cd26a6c2b73bdf3b6446b05e567", "staged_path": "results/historical-al-curves/v1/b7_math/al_curve.csv", "staged_sha256": "ea0b445042ac886293657641a64bb039fd6e2cd26a6c2b73bdf3b6446b05e567", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c0_code/al_curve.csv", "source_sha256": "4c7950c6ca3dbe06d5674d2568b5b183b56de92bf4fa76230af5c5d045e1eabd", "staged_path": "results/historical-al-curves/v1/c0_code/al_curve.csv", "staged_sha256": "4c7950c6ca3dbe06d5674d2568b5b183b56de92bf4fa76230af5c5d045e1eabd", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/c0_creative_writing/al_curve.csv", "source_sha256": "b74b5f158b2f61a23d899aec307d8615f8a33d158032b2a896804b0bf1450514", "staged_path": "results/historical-al-curves/v1/c0_creative_writing/al_curve.csv", "staged_sha256": "b74b5f158b2f61a23d899aec307d8615f8a33d158032b2a896804b0bf1450514", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/c0_factual_qa/al_curve.csv", "source_sha256": "b4e8a4ec59a51f14f79275cb1341280656a6bee50653fe497c88680af5acbcee", "staged_path": "results/historical-al-curves/v1/c0_factual_qa/al_curve.csv", "staged_sha256": "b4e8a4ec59a51f14f79275cb1341280656a6bee50653fe497c88680af5acbcee", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 79, "source_path": "experiments/dflash/eval_csv/c0_general/al_curve.csv", "source_sha256": "0ee93e536ed44dbcfab2d08f8f8a76e6590b4f35cc8fc1a88cb58248700b2a78", "staged_path": "results/historical-al-curves/v1/c0_general/al_curve.csv", "staged_sha256": "0ee93e536ed44dbcfab2d08f8f8a76e6590b4f35cc8fc1a88cb58248700b2a78", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c0_math/al_curve.csv", "source_sha256": "607c0206b4b6e6985297e22ee36d60fc8d315d5275b5b1d10546d2ec5de1b44e", "staged_path": "results/historical-al-curves/v1/c0_math/al_curve.csv", "staged_sha256": "607c0206b4b6e6985297e22ee36d60fc8d315d5275b5b1d10546d2ec5de1b44e", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 73, "source_path": "experiments/dflash/eval_csv/c1_code/al_curve.csv", "source_sha256": "d7b3403ff3fe7e8cc6e119f046911c339bc297fb9d6484ea667326c5b4993f52", "staged_path": "results/historical-al-curves/v1/c1_code/al_curve.csv", "staged_sha256": "d7b3403ff3fe7e8cc6e119f046911c339bc297fb9d6484ea667326c5b4993f52", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/c1_creative_writing/al_curve.csv", "source_sha256": "954ff064cf4cec980f9439cd5c7434272ffc93ecd38f0eb7fba1c7c08c1fa518", "staged_path": "results/historical-al-curves/v1/c1_creative_writing/al_curve.csv", "staged_sha256": "954ff064cf4cec980f9439cd5c7434272ffc93ecd38f0eb7fba1c7c08c1fa518", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 77, "source_path": "experiments/dflash/eval_csv/c1_factual_qa/al_curve.csv", "source_sha256": "db592b66feeeb1b807c0cd192b474eb6beca4b3be00d1cf9bd1faafd6175b47a", "staged_path": "results/historical-al-curves/v1/c1_factual_qa/al_curve.csv", "staged_sha256": "db592b66feeeb1b807c0cd192b474eb6beca4b3be00d1cf9bd1faafd6175b47a", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 78, "source_path": "experiments/dflash/eval_csv/c1_general/al_curve.csv", "source_sha256": "803576dbe5083e7b3a487a9780cb06bb58c63e82601156afe931fd3e04d7155d", "staged_path": "results/historical-al-curves/v1/c1_general/al_curve.csv", "staged_sha256": "803576dbe5083e7b3a487a9780cb06bb58c63e82601156afe931fd3e04d7155d", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c1_math/al_curve.csv", "source_sha256": "eca7f51ff784366375d5e29799fc5bf316a9ef803812d14e22c066f508c1cb35", "staged_path": "results/historical-al-curves/v1/c1_math/al_curve.csv", "staged_sha256": "eca7f51ff784366375d5e29799fc5bf316a9ef803812d14e22c066f508c1cb35", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c2_code/al_curve.csv", "source_sha256": "65685989c83c8cde6179b3baf8725a231c5d020e326accc2f4ccb65db4ebecb5", "staged_path": "results/historical-al-curves/v1/c2_code/al_curve.csv", "staged_sha256": "65685989c83c8cde6179b3baf8725a231c5d020e326accc2f4ccb65db4ebecb5", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/c2_creative_writing/al_curve.csv", "source_sha256": "868681f811ac8721811740d3c504f35bf37d1560d67d045c86e68023b7ea3dcd", "staged_path": "results/historical-al-curves/v1/c2_creative_writing/al_curve.csv", "staged_sha256": "868681f811ac8721811740d3c504f35bf37d1560d67d045c86e68023b7ea3dcd", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/c2_factual_qa/al_curve.csv", "source_sha256": "d5b2c7f25d9bcfb1ee072c5529e8cdfa8122de16cf8ae5b25601a27ea4b6b684", "staged_path": "results/historical-al-curves/v1/c2_factual_qa/al_curve.csv", "staged_sha256": "d5b2c7f25d9bcfb1ee072c5529e8cdfa8122de16cf8ae5b25601a27ea4b6b684", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/c2_general/al_curve.csv", "source_sha256": "211aca725b163fc8c6db2956352f254e03005204551d0bce4fe2c11fc24166ca", "staged_path": "results/historical-al-curves/v1/c2_general/al_curve.csv", "staged_sha256": "211aca725b163fc8c6db2956352f254e03005204551d0bce4fe2c11fc24166ca", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c2_math/al_curve.csv", "source_sha256": "ba923a05b754e4009f41f452da060bf128b4af1d97b5d60c69cb0adf970964b4", "staged_path": "results/historical-al-curves/v1/c2_math/al_curve.csv", "staged_sha256": "ba923a05b754e4009f41f452da060bf128b4af1d97b5d60c69cb0adf970964b4", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c3_code/al_curve.csv", "source_sha256": "7a7032f0fdb54aff1bb68456bcc4619107572d9e1c7be9f04653dffe903e2c31", "staged_path": "results/historical-al-curves/v1/c3_code/al_curve.csv", "staged_sha256": "7a7032f0fdb54aff1bb68456bcc4619107572d9e1c7be9f04653dffe903e2c31", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/c3_creative_writing/al_curve.csv", "source_sha256": "dd88bd7aa0c7d70467c88eba8ea31c26f4db5aa82c6c485f723f1bc6a8a47f7a", "staged_path": "results/historical-al-curves/v1/c3_creative_writing/al_curve.csv", "staged_sha256": "dd88bd7aa0c7d70467c88eba8ea31c26f4db5aa82c6c485f723f1bc6a8a47f7a", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/c3_factual_qa/al_curve.csv", "source_sha256": "8790230d20b4996547849d1fee887ccf70a036fb69c3377c93ef237b74e9576f", "staged_path": "results/historical-al-curves/v1/c3_factual_qa/al_curve.csv", "staged_sha256": "8790230d20b4996547849d1fee887ccf70a036fb69c3377c93ef237b74e9576f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 79, "source_path": "experiments/dflash/eval_csv/c3_general/al_curve.csv", "source_sha256": "d1586a3cf87eee78a83d6fdac95a70203c4b18fa56f371b1f3f1b801fe1c27b1", "staged_path": "results/historical-al-curves/v1/c3_general/al_curve.csv", "staged_sha256": "d1586a3cf87eee78a83d6fdac95a70203c4b18fa56f371b1f3f1b801fe1c27b1", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 71, "source_path": "experiments/dflash/eval_csv/c3_math/al_curve.csv", "source_sha256": "4ad7ce036023a5644f713848b56b428fd78ba427e77ab2aea8039105c12cc5de", "staged_path": "results/historical-al-curves/v1/c3_math/al_curve.csv", "staged_sha256": "4ad7ce036023a5644f713848b56b428fd78ba427e77ab2aea8039105c12cc5de", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/c4_code/al_curve.csv", "source_sha256": "7155ecbb5af764aff597a5071312c4fbe7ea3b9678d438e53a84db5994d544c3", "staged_path": "results/historical-al-curves/v1/c4_code/al_curve.csv", "staged_sha256": "7155ecbb5af764aff597a5071312c4fbe7ea3b9678d438e53a84db5994d544c3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/c4_creative_writing/al_curve.csv", "source_sha256": "7a9502461a40264469134c1602dce11605d746fb362d7946826addfabe04ee9c", "staged_path": "results/historical-al-curves/v1/c4_creative_writing/al_curve.csv", "staged_sha256": "7a9502461a40264469134c1602dce11605d746fb362d7946826addfabe04ee9c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/c4_factual_qa/al_curve.csv", "source_sha256": "deec1ade9fbb0ca6120706265aebad6ed2ac5dfd1303a4235276dfe4dd86af9a", "staged_path": "results/historical-al-curves/v1/c4_factual_qa/al_curve.csv", "staged_sha256": "deec1ade9fbb0ca6120706265aebad6ed2ac5dfd1303a4235276dfe4dd86af9a", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/c4_general/al_curve.csv", "source_sha256": "e92a4cdf7a876ca2100795d03735b95741af6acd8346c11fb2baca421f9b8b13", "staged_path": "results/historical-al-curves/v1/c4_general/al_curve.csv", "staged_sha256": "e92a4cdf7a876ca2100795d03735b95741af6acd8346c11fb2baca421f9b8b13", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c4_math/al_curve.csv", "source_sha256": "e42b09abd0da17dadf2ec15a8f3a97ebd8abe052ed1cd28971fd35c4d82979b7", "staged_path": "results/historical-al-curves/v1/c4_math/al_curve.csv", "staged_sha256": "e42b09abd0da17dadf2ec15a8f3a97ebd8abe052ed1cd28971fd35c4d82979b7", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/c5_code/al_curve.csv", "source_sha256": "3b5f8477eb60ba034b09b80033824d38bc061efb987392bba95ebf82f84085b2", "staged_path": "results/historical-al-curves/v1/c5_code/al_curve.csv", "staged_sha256": "3b5f8477eb60ba034b09b80033824d38bc061efb987392bba95ebf82f84085b2", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 97, "source_path": "experiments/dflash/eval_csv/c5_creative_writing/al_curve.csv", "source_sha256": "a7c05f9e0de662069c4c327511e5a44d8c5c2600a8a3d327d23a18e91f29fc75", "staged_path": "results/historical-al-curves/v1/c5_creative_writing/al_curve.csv", "staged_sha256": "a7c05f9e0de662069c4c327511e5a44d8c5c2600a8a3d327d23a18e91f29fc75", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/c5_factual_qa/al_curve.csv", "source_sha256": "e38286d943199cec158517b54a799287c37c50d78e996d30ec13c8099c5cb080", "staged_path": "results/historical-al-curves/v1/c5_factual_qa/al_curve.csv", "staged_sha256": "e38286d943199cec158517b54a799287c37c50d78e996d30ec13c8099c5cb080", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/c5_general/al_curve.csv", "source_sha256": "160423ad74f8f5bf5924b853e143138763795e3e7f0432729ebf07426d708c56", "staged_path": "results/historical-al-curves/v1/c5_general/al_curve.csv", "staged_sha256": "160423ad74f8f5bf5924b853e143138763795e3e7f0432729ebf07426d708c56", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c5_math/al_curve.csv", "source_sha256": "093d79e141db17e3f2da8411b754057585377e77dbe2ca4b0c021c18721233ce", "staged_path": "results/historical-al-curves/v1/c5_math/al_curve.csv", "staged_sha256": "093d79e141db17e3f2da8411b754057585377e77dbe2ca4b0c021c18721233ce", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c6_code/al_curve.csv", "source_sha256": "efa094eefd03080e2d1612e2780cabc4b4fe82102bae3a70d685bf1fd9a5bdce", "staged_path": "results/historical-al-curves/v1/c6_code/al_curve.csv", "staged_sha256": "efa094eefd03080e2d1612e2780cabc4b4fe82102bae3a70d685bf1fd9a5bdce", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/c6_creative_writing/al_curve.csv", "source_sha256": "990c56cc487367e9e418bbeff9132144433f082632b7416e2d22dd36f003c689", "staged_path": "results/historical-al-curves/v1/c6_creative_writing/al_curve.csv", "staged_sha256": "990c56cc487367e9e418bbeff9132144433f082632b7416e2d22dd36f003c689", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/c6_factual_qa/al_curve.csv", "source_sha256": "e2ce13345d49e2960fa9e52649675e0a1c1af4b61f59e5d13a2375fb806263ee", "staged_path": "results/historical-al-curves/v1/c6_factual_qa/al_curve.csv", "staged_sha256": "e2ce13345d49e2960fa9e52649675e0a1c1af4b61f59e5d13a2375fb806263ee", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/c6_general/al_curve.csv", "source_sha256": "5d94f62ca87701b51af5ef930fe63be1df7958e85773ccd402edeacbf59a25b1", "staged_path": "results/historical-al-curves/v1/c6_general/al_curve.csv", "staged_sha256": "5d94f62ca87701b51af5ef930fe63be1df7958e85773ccd402edeacbf59a25b1", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c6_math/al_curve.csv", "source_sha256": "4fc731227fb3cc66f90a5152759c216faada25a65fc332d8148541b9f1c4bac0", "staged_path": "results/historical-al-curves/v1/c6_math/al_curve.csv", "staged_sha256": "4fc731227fb3cc66f90a5152759c216faada25a65fc332d8148541b9f1c4bac0", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/c7_code/al_curve.csv", "source_sha256": "4e8ea13685b72455b24ab172fa68d71743bc71a6d7ed7462fe30e36ee16a08d7", "staged_path": "results/historical-al-curves/v1/c7_code/al_curve.csv", "staged_sha256": "4e8ea13685b72455b24ab172fa68d71743bc71a6d7ed7462fe30e36ee16a08d7", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/c7_creative_writing/al_curve.csv", "source_sha256": "307586723ae39dda90995aded367c77aa254e3820a2ded9f42f2c62ef83f463f", "staged_path": "results/historical-al-curves/v1/c7_creative_writing/al_curve.csv", "staged_sha256": "307586723ae39dda90995aded367c77aa254e3820a2ded9f42f2c62ef83f463f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/c7_factual_qa/al_curve.csv", "source_sha256": "ecff8fdc0258485e27b1f4ebf11449d4345b6f12cae0c7ef4ecbd4c0d3705167", "staged_path": "results/historical-al-curves/v1/c7_factual_qa/al_curve.csv", "staged_sha256": "ecff8fdc0258485e27b1f4ebf11449d4345b6f12cae0c7ef4ecbd4c0d3705167", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/c7_general/al_curve.csv", "source_sha256": "927daac26248cf56a8e7f09bd8c6b4a19a1f841cfb6b64d3d8643b5c0e017832", "staged_path": "results/historical-al-curves/v1/c7_general/al_curve.csv", "staged_sha256": "927daac26248cf56a8e7f09bd8c6b4a19a1f841cfb6b64d3d8643b5c0e017832", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/c7_math/al_curve.csv", "source_sha256": "e4af3170b25fc05d0ad872adb322836852e15d0d40fca7b7fd4f7f7b27b8dd62", "staged_path": "results/historical-al-curves/v1/c7_math/al_curve.csv", "staged_sha256": "e4af3170b25fc05d0ad872adb322836852e15d0d40fca7b7fd4f7f7b27b8dd62", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/d0_code/al_curve.csv", "source_sha256": "1789f2c326487f364f3c14bc9ef4773d2f5e6a3ddaee65cb8b1c2847f6116d47", "staged_path": "results/historical-al-curves/v1/d0_code/al_curve.csv", "staged_sha256": "1789f2c326487f364f3c14bc9ef4773d2f5e6a3ddaee65cb8b1c2847f6116d47", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 72, "source_path": "experiments/dflash/eval_csv/d0_code_bench/al_curve.csv", "source_sha256": "cc4024858d44b0252ff081901cb2d710e83257679835f6d40ff8c8744db33602", "staged_path": "results/historical-al-curves/v1/d0_code_bench/al_curve.csv", "staged_sha256": "cc4024858d44b0252ff081901cb2d710e83257679835f6d40ff8c8744db33602", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/d0_creative_writing/al_curve.csv", "source_sha256": "02c34be9c790947677503f07b3873849e14458c3cf18efed4717be917dadddab", "staged_path": "results/historical-al-curves/v1/d0_creative_writing/al_curve.csv", "staged_sha256": "02c34be9c790947677503f07b3873849e14458c3cf18efed4717be917dadddab", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 85, "source_path": "experiments/dflash/eval_csv/d0_creative_writing_bench/al_curve.csv", "source_sha256": "3eb6cfcd4fa4a755eb9d2f9a9c04ad4a82341a5eee734be5922d0a3530568c30", "staged_path": "results/historical-al-curves/v1/d0_creative_writing_bench/al_curve.csv", "staged_sha256": "3eb6cfcd4fa4a755eb9d2f9a9c04ad4a82341a5eee734be5922d0a3530568c30", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/d0_factual_qa/al_curve.csv", "source_sha256": "2140012b6c2b0f52b2aada13cfad0cdae99afeb8b7936281c65e197c2f8b0fac", "staged_path": "results/historical-al-curves/v1/d0_factual_qa/al_curve.csv", "staged_sha256": "2140012b6c2b0f52b2aada13cfad0cdae99afeb8b7936281c65e197c2f8b0fac", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 78, "source_path": "experiments/dflash/eval_csv/d0_factual_qa_bench/al_curve.csv", "source_sha256": "b88004349fe0af1749dcfd00d3d1719c63079daab247bd961224ef7d5355eb20", "staged_path": "results/historical-al-curves/v1/d0_factual_qa_bench/al_curve.csv", "staged_sha256": "b88004349fe0af1749dcfd00d3d1719c63079daab247bd961224ef7d5355eb20", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/d0_general/al_curve.csv", "source_sha256": "511471c17e2d15eedbd3481e7a9771170fccfdeeec09916974e3c1633793ee86", "staged_path": "results/historical-al-curves/v1/d0_general/al_curve.csv", "staged_sha256": "511471c17e2d15eedbd3481e7a9771170fccfdeeec09916974e3c1633793ee86", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 76, "source_path": "experiments/dflash/eval_csv/d0_general_bench/al_curve.csv", "source_sha256": "898934747a054fa24b146c14a5f2aba09d61152cc5c38a902cf4f56ff5af1d42", "staged_path": "results/historical-al-curves/v1/d0_general_bench/al_curve.csv", "staged_sha256": "898934747a054fa24b146c14a5f2aba09d61152cc5c38a902cf4f56ff5af1d42", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 73, "source_path": "experiments/dflash/eval_csv/d0_math/al_curve.csv", "source_sha256": "5c4e2ae4caf4bef8303353cde38d98055123220e68f69c6896d65296edb81eb4", "staged_path": "results/historical-al-curves/v1/d0_math/al_curve.csv", "staged_sha256": "5c4e2ae4caf4bef8303353cde38d98055123220e68f69c6896d65296edb81eb4", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 73, "source_path": "experiments/dflash/eval_csv/d0_math_bench/al_curve.csv", "source_sha256": "1e361705ccdbb2fa0507cfc3beffebddd392d36038efa3e8dc2542d281777545", "staged_path": "results/historical-al-curves/v1/d0_math_bench/al_curve.csv", "staged_sha256": "1e361705ccdbb2fa0507cfc3beffebddd392d36038efa3e8dc2542d281777545", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/d1_code/al_curve.csv", "source_sha256": "ebf1b0e2d4ec3da18de3011f029c029a81b8b14407c9b1d6a78333af3f07f433", "staged_path": "results/historical-al-curves/v1/d1_code/al_curve.csv", "staged_sha256": "ebf1b0e2d4ec3da18de3011f029c029a81b8b14407c9b1d6a78333af3f07f433", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/d1_creative_writing/al_curve.csv", "source_sha256": "8912e2738a3460457ee2b1968dd3dacf93e7fdaa6b22cc064359777f5f5ccf01", "staged_path": "results/historical-al-curves/v1/d1_creative_writing/al_curve.csv", "staged_sha256": "8912e2738a3460457ee2b1968dd3dacf93e7fdaa6b22cc064359777f5f5ccf01", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/d1_factual_qa/al_curve.csv", "source_sha256": "ed8eed33bff28b02c080d46b998217298a68e630abc23e732cc47251acadb2ab", "staged_path": "results/historical-al-curves/v1/d1_factual_qa/al_curve.csv", "staged_sha256": "ed8eed33bff28b02c080d46b998217298a68e630abc23e732cc47251acadb2ab", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/d1_general/al_curve.csv", "source_sha256": "91824d146176f24973bfa6fbfe8a6d55249ac4881d1d5a23935c13f1d2c51618", "staged_path": "results/historical-al-curves/v1/d1_general/al_curve.csv", "staged_sha256": "91824d146176f24973bfa6fbfe8a6d55249ac4881d1d5a23935c13f1d2c51618", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/d1_math/al_curve.csv", "source_sha256": "8ccaf221637cc8cf7631ad7d9dd260adcf9f09520eff899ee72c5867f44f1714", "staged_path": "results/historical-al-curves/v1/d1_math/al_curve.csv", "staged_sha256": "8ccaf221637cc8cf7631ad7d9dd260adcf9f09520eff899ee72c5867f44f1714", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/d2_code/al_curve.csv", "source_sha256": "2ba1138d82f05ed5fd7bc45186cff4351328324ad93608d2c25c20ddc8fe0b2d", "staged_path": "results/historical-al-curves/v1/d2_code/al_curve.csv", "staged_sha256": "2ba1138d82f05ed5fd7bc45186cff4351328324ad93608d2c25c20ddc8fe0b2d", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/d2_creative_writing/al_curve.csv", "source_sha256": "9c0dab3f0616fb762d43aeecab1e19d292d92ba3b72e74efc6ee3c37d2547dcf", "staged_path": "results/historical-al-curves/v1/d2_creative_writing/al_curve.csv", "staged_sha256": "9c0dab3f0616fb762d43aeecab1e19d292d92ba3b72e74efc6ee3c37d2547dcf", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/d2_factual_qa/al_curve.csv", "source_sha256": "c63e517f6e5970d1b2208a455e50073b5fd531cfd9484b70890577a72baa74bf", "staged_path": "results/historical-al-curves/v1/d2_factual_qa/al_curve.csv", "staged_sha256": "c63e517f6e5970d1b2208a455e50073b5fd531cfd9484b70890577a72baa74bf", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/d2_general/al_curve.csv", "source_sha256": "849b26bbb6d4cb5ec1b2ae207745d0c8cb5433b274404e116bb4a41bd9649f96", "staged_path": "results/historical-al-curves/v1/d2_general/al_curve.csv", "staged_sha256": "849b26bbb6d4cb5ec1b2ae207745d0c8cb5433b274404e116bb4a41bd9649f96", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/d2_math/al_curve.csv", "source_sha256": "9a6dd3a37a3daf3368a2262863b82ae34e47192e25d16ab763b4dbd39f7a5799", "staged_path": "results/historical-al-curves/v1/d2_math/al_curve.csv", "staged_sha256": "9a6dd3a37a3daf3368a2262863b82ae34e47192e25d16ab763b4dbd39f7a5799", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/d3_code/al_curve.csv", "source_sha256": "48877d7e509598be3940beed6654af434efa1de2d9e4f46eb42352ddeb22c8ec", "staged_path": "results/historical-al-curves/v1/d3_code/al_curve.csv", "staged_sha256": "48877d7e509598be3940beed6654af434efa1de2d9e4f46eb42352ddeb22c8ec", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/d3_creative_writing/al_curve.csv", "source_sha256": "2dc5e5d683438c79532dfabbe065e694a6b9603fa5ef7e73648eee7ad0ec7138", "staged_path": "results/historical-al-curves/v1/d3_creative_writing/al_curve.csv", "staged_sha256": "2dc5e5d683438c79532dfabbe065e694a6b9603fa5ef7e73648eee7ad0ec7138", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/d3_factual_qa/al_curve.csv", "source_sha256": "e2e5ed61a2ed83cd6e09030ce52692106a8ae7ebdbe1c18bdf58fc0abe150845", "staged_path": "results/historical-al-curves/v1/d3_factual_qa/al_curve.csv", "staged_sha256": "e2e5ed61a2ed83cd6e09030ce52692106a8ae7ebdbe1c18bdf58fc0abe150845", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/d3_general/al_curve.csv", "source_sha256": "f080acc44e7d4ddc6a7cc253a722d9cfb73b531e235e0605101f951fed43e47c", "staged_path": "results/historical-al-curves/v1/d3_general/al_curve.csv", "staged_sha256": "f080acc44e7d4ddc6a7cc253a722d9cfb73b531e235e0605101f951fed43e47c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/d3_math/al_curve.csv", "source_sha256": "ba98a34a8360433870e3e12475fc275a7ab6d03bcc9b94029e7fa81a8dd97f2c", "staged_path": "results/historical-al-curves/v1/d3_math/al_curve.csv", "staged_sha256": "ba98a34a8360433870e3e12475fc275a7ab6d03bcc9b94029e7fa81a8dd97f2c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/d4_code/al_curve.csv", "source_sha256": "06430bbb7a1dde00cb11eae287cf41212f0c4270fab85d6a7b4ea9cdfcc6376e", "staged_path": "results/historical-al-curves/v1/d4_code/al_curve.csv", "staged_sha256": "06430bbb7a1dde00cb11eae287cf41212f0c4270fab85d6a7b4ea9cdfcc6376e", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/d4_creative_writing/al_curve.csv", "source_sha256": "138322d96b562a197c3dfce4be96877a63d5c60da9299baf9a4e1462e0457057", "staged_path": "results/historical-al-curves/v1/d4_creative_writing/al_curve.csv", "staged_sha256": "138322d96b562a197c3dfce4be96877a63d5c60da9299baf9a4e1462e0457057", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/d5_code/al_curve.csv", "source_sha256": "f6a4e0d1aabcd786e184e3845c2d8f07afff7938c8faacc6a6e5831024300e55", "staged_path": "results/historical-al-curves/v1/d5_code/al_curve.csv", "staged_sha256": "f6a4e0d1aabcd786e184e3845c2d8f07afff7938c8faacc6a6e5831024300e55", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/d5_creative_writing/al_curve.csv", "source_sha256": "544e7047314513d17b726f10c4a47e3069d3d48eddb779cde9bef3f3aa9acb98", "staged_path": "results/historical-al-curves/v1/d5_creative_writing/al_curve.csv", "staged_sha256": "544e7047314513d17b726f10c4a47e3069d3d48eddb779cde9bef3f3aa9acb98", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/d5_factual_qa/al_curve.csv", "source_sha256": "2c4c0da28aec12ee646a01c36e54e226f6f2f7c2598daa589e24083b753178dc", "staged_path": "results/historical-al-curves/v1/d5_factual_qa/al_curve.csv", "staged_sha256": "2c4c0da28aec12ee646a01c36e54e226f6f2f7c2598daa589e24083b753178dc", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/d5_general/al_curve.csv", "source_sha256": "db3224fa3489c952487463cb049c202e60a01ef6665d2d5a2aecc01d44c2985b", "staged_path": "results/historical-al-curves/v1/d5_general/al_curve.csv", "staged_sha256": "db3224fa3489c952487463cb049c202e60a01ef6665d2d5a2aecc01d44c2985b", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 73, "source_path": "experiments/dflash/eval_csv/d5_math/al_curve.csv", "source_sha256": "e492b9a6a7e2fa83288c02c0f450739f2d30d124d6e05aa8cce23225792fdafe", "staged_path": "results/historical-al-curves/v1/d5_math/al_curve.csv", "staged_sha256": "e492b9a6a7e2fa83288c02c0f450739f2d30d124d6e05aa8cce23225792fdafe", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/d6_code/al_curve.csv", "source_sha256": "d6fe8d0beec5432eab9c8fe8a9b4fcbb7b08effc638f07bc8d56dd73cf18b1bb", "staged_path": "results/historical-al-curves/v1/d6_code/al_curve.csv", "staged_sha256": "d6fe8d0beec5432eab9c8fe8a9b4fcbb7b08effc638f07bc8d56dd73cf18b1bb", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/d6_creative_writing/al_curve.csv", "source_sha256": "dcc8681fc1864c9af927f1acb89ca6e18cf2b8bb16738eb500779cdd79ef1321", "staged_path": "results/historical-al-curves/v1/d6_creative_writing/al_curve.csv", "staged_sha256": "dcc8681fc1864c9af927f1acb89ca6e18cf2b8bb16738eb500779cdd79ef1321", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 86, "source_path": "experiments/dflash/eval_csv/d6_factual_qa/al_curve.csv", "source_sha256": "ec1c8abcab0e73ceb72b7aeeaa56cacc483e032fe97bdb27b13096f659f933a4", "staged_path": "results/historical-al-curves/v1/d6_factual_qa/al_curve.csv", "staged_sha256": "ec1c8abcab0e73ceb72b7aeeaa56cacc483e032fe97bdb27b13096f659f933a4", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 79, "source_path": "experiments/dflash/eval_csv/d6_general/al_curve.csv", "source_sha256": "38416012f7b76ee350ed97bb6efe3f2b3454a430a1a588e3a1007035804f26fc", "staged_path": "results/historical-al-curves/v1/d6_general/al_curve.csv", "staged_sha256": "38416012f7b76ee350ed97bb6efe3f2b3454a430a1a588e3a1007035804f26fc", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/d6_math/al_curve.csv", "source_sha256": "a9ed8256345309365b7cd216a0c587ab31c6e66d76eb9790c4b66dafd2e96859", "staged_path": "results/historical-al-curves/v1/d6_math/al_curve.csv", "staged_sha256": "a9ed8256345309365b7cd216a0c587ab31c6e66d76eb9790c4b66dafd2e96859", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/d7_code/al_curve.csv", "source_sha256": "cee9296d23831df5ee2411bb8a5f95c905796418fe20293c1a7c92981cdb99fe", "staged_path": "results/historical-al-curves/v1/d7_code/al_curve.csv", "staged_sha256": "cee9296d23831df5ee2411bb8a5f95c905796418fe20293c1a7c92981cdb99fe", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/d7_creative_writing/al_curve.csv", "source_sha256": "1dc7a8e5e0b04f920984b60ff2203bcf6dac7f48a5f331ffe0189bd893387d74", "staged_path": "results/historical-al-curves/v1/d7_creative_writing/al_curve.csv", "staged_sha256": "1dc7a8e5e0b04f920984b60ff2203bcf6dac7f48a5f331ffe0189bd893387d74", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/d7_factual_qa/al_curve.csv", "source_sha256": "701dbf445877534d1817c9a7132023654f10ea50e4878c3aebd98a79614e60f5", "staged_path": "results/historical-al-curves/v1/d7_factual_qa/al_curve.csv", "staged_sha256": "701dbf445877534d1817c9a7132023654f10ea50e4878c3aebd98a79614e60f5", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/d7_general/al_curve.csv", "source_sha256": "bb45782350192168b00e368c3919d1a133096e2ba93b6c583f90ea903ab4a908", "staged_path": "results/historical-al-curves/v1/d7_general/al_curve.csv", "staged_sha256": "bb45782350192168b00e368c3919d1a133096e2ba93b6c583f90ea903ab4a908", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/d7_math/al_curve.csv", "source_sha256": "ca8299b3e74ed2a8b1831f7ecaf2cfecff9c8931f1396da632354b4b6e93a4d9", "staged_path": "results/historical-al-curves/v1/d7_math/al_curve.csv", "staged_sha256": "ca8299b3e74ed2a8b1831f7ecaf2cfecff9c8931f1396da632354b4b6e93a4d9", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 416, "source_path": "experiments/dflash/eval_csv/gendense_l0/al_curve.csv", "source_sha256": "b5f3417a3a63c110dfa9d0bd7af497940873925f4f2ac969816034d64b54908c", "staged_path": "results/historical-al-curves/v1/gendense_l0/al_curve.csv", "staged_sha256": "b5f3417a3a63c110dfa9d0bd7af497940873925f4f2ac969816034d64b54908c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 418, "source_path": "experiments/dflash/eval_csv/gendense_l1/al_curve.csv", "source_sha256": "cc92b9d1cef97edf64e31bfabe6ef9b2dcfcc343d7f2c3fc492187adf72fce57", "staged_path": "results/historical-al-curves/v1/gendense_l1/al_curve.csv", "staged_sha256": "cc92b9d1cef97edf64e31bfabe6ef9b2dcfcc343d7f2c3fc492187adf72fce57", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 418, "source_path": "experiments/dflash/eval_csv/gendense_l2/al_curve.csv", "source_sha256": "76d88265db58b5a24a79096ed0dc6dc8ca2e5e24024a15c8777e38e6e0895eee", "staged_path": "results/historical-al-curves/v1/gendense_l2/al_curve.csv", "staged_sha256": "76d88265db58b5a24a79096ed0dc6dc8ca2e5e24024a15c8777e38e6e0895eee", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 417, "source_path": "experiments/dflash/eval_csv/gendense_l3/al_curve.csv", "source_sha256": "6e29784a0f479940051139ec7bc3b281c5d747f8169770daf008a05b64e72414", "staged_path": "results/historical-al-curves/v1/gendense_l3/al_curve.csv", "staged_sha256": "6e29784a0f479940051139ec7bc3b281c5d747f8169770daf008a05b64e72414", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 363, "source_path": "experiments/dflash/eval_csv/gendense_l4/al_curve.csv", "source_sha256": "74378cdfcbbdb0342d963b8368350ca0a2dcaf7a9e331bbedd6b67bda9e313a5", "staged_path": "results/historical-al-curves/v1/gendense_l4/al_curve.csv", "staged_sha256": "74378cdfcbbdb0342d963b8368350ca0a2dcaf7a9e331bbedd6b67bda9e313a5", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 362, "source_path": "experiments/dflash/eval_csv/gendense_l5/al_curve.csv", "source_sha256": "cb8e712bf404f41ce40f3329276dcef71a9e05852fe4c230f2a21a23a4cf348c", "staged_path": "results/historical-al-curves/v1/gendense_l5/al_curve.csv", "staged_sha256": "cb8e712bf404f41ce40f3329276dcef71a9e05852fe4c230f2a21a23a4cf348c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 361, "source_path": "experiments/dflash/eval_csv/gendense_l6/al_curve.csv", "source_sha256": "7cd3d94ff8987950768e4722014f3956f2af4114d1bdb428361028603b6d1088", "staged_path": "results/historical-al-curves/v1/gendense_l6/al_curve.csv", "staged_sha256": "7cd3d94ff8987950768e4722014f3956f2af4114d1bdb428361028603b6d1088", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 364, "source_path": "experiments/dflash/eval_csv/gendense_l7/al_curve.csv", "source_sha256": "e36b37efcd21b27b438bf1015266172c0804f369b48d64c243ab70cc9ab34867", "staged_path": "results/historical-al-curves/v1/gendense_l7/al_curve.csv", "staged_sha256": "e36b37efcd21b27b438bf1015266172c0804f369b48d64c243ab70cc9ab34867", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 94, "source_path": "experiments/dflash/eval_csv/genep4_code/al_curve.csv", "source_sha256": "edf093e7b7be16957278d8c961197b9e913be5b5ac0678df48bcd54e8c963fbf", "staged_path": "results/historical-al-curves/v1/genep4_code/al_curve.csv", "staged_sha256": "edf093e7b7be16957278d8c961197b9e913be5b5ac0678df48bcd54e8c963fbf", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 94, "source_path": "experiments/dflash/eval_csv/genep4_math/al_curve.csv", "source_sha256": "644c907eb90c2ed4790bf0805df65af178fa4348076e6b7a5258bbe7bb5eb7e4", "staged_path": "results/historical-al-curves/v1/genep4_math/al_curve.csv", "staged_sha256": "644c907eb90c2ed4790bf0805df65af178fa4348076e6b7a5258bbe7bb5eb7e4", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 298, "source_path": "experiments/dflash/eval_csv/merge89/al_curve.csv", "source_sha256": "f575b47ecd4d248d4f4c0f4a2e6e685ee60fc9aff335f0eae2e64d06d503ff17", "staged_path": "results/historical-al-curves/v1/merge89/al_curve.csv", "staged_sha256": "f575b47ecd4d248d4f4c0f4a2e6e685ee60fc9aff335f0eae2e64d06d503ff17", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 695, "source_path": "experiments/dflash/eval_csv/mv2c89/al_curve.csv", "source_sha256": "ffa68e561efb0ff335945d7f5ce2c4ce884c1cea35fbecc0acfc8f0a3a7f5a1c", "staged_path": "results/historical-al-curves/v1/mv2c89/al_curve.csv", "staged_sha256": "ffa68e561efb0ff335945d7f5ce2c4ce884c1cea35fbecc0acfc8f0a3a7f5a1c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 292, "source_path": "experiments/dflash/eval_csv/mv3x89/al_curve.csv", "source_sha256": "acfb67b4fa59a0eb7a175f2c110e8b609eb2ff530283f5ad770e949b3d15d595", "staged_path": "results/historical-al-curves/v1/mv3x89/al_curve.csv", "staged_sha256": "acfb67b4fa59a0eb7a175f2c110e8b609eb2ff530283f5ad770e949b3d15d595", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/mx_code_code/al_curve.csv", "source_sha256": "62fe8604350ac825389c66aa3282b9c5076cfac2cbd33ce35f98222c8995edfd", "staged_path": "results/historical-al-curves/v1/mx_code_code/al_curve.csv", "staged_sha256": "62fe8604350ac825389c66aa3282b9c5076cfac2cbd33ce35f98222c8995edfd", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/mx_code_creative_writing/al_curve.csv", "source_sha256": "471ed1c8cb6266ce5462dd90f4fd767ff224b794b6431b51207b3db10e953d86", "staged_path": "results/historical-al-curves/v1/mx_code_creative_writing/al_curve.csv", "staged_sha256": "471ed1c8cb6266ce5462dd90f4fd767ff224b794b6431b51207b3db10e953d86", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/mx_code_factual_qa/al_curve.csv", "source_sha256": "a0ac4e1243a51d3a5ab2aaf79595f8e89836f1f43210073c32fa36530bef3638", "staged_path": "results/historical-al-curves/v1/mx_code_factual_qa/al_curve.csv", "staged_sha256": "a0ac4e1243a51d3a5ab2aaf79595f8e89836f1f43210073c32fa36530bef3638", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 77, "source_path": "experiments/dflash/eval_csv/mx_code_general/al_curve.csv", "source_sha256": "75ff837f1749d0f095ca7ba242cf6d00d7817b27049efa3c2cc4b51f9c5b4a9a", "staged_path": "results/historical-al-curves/v1/mx_code_general/al_curve.csv", "staged_sha256": "75ff837f1749d0f095ca7ba242cf6d00d7817b27049efa3c2cc4b51f9c5b4a9a", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 74, "source_path": "experiments/dflash/eval_csv/mx_code_math/al_curve.csv", "source_sha256": "27e2d44d0e1d64adbdc9dbb02f8fdfd5849a1acb65e886bca1b663dc947b1df2", "staged_path": "results/historical-al-curves/v1/mx_code_math/al_curve.csv", "staged_sha256": "27e2d44d0e1d64adbdc9dbb02f8fdfd5849a1acb65e886bca1b663dc947b1df2", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/mx_creative_writing_code/al_curve.csv", "source_sha256": "193bbf84b8b4802b7ca15e36ff505558842da29023f3f450441b771e3244bdc0", "staged_path": "results/historical-al-curves/v1/mx_creative_writing_code/al_curve.csv", "staged_sha256": "193bbf84b8b4802b7ca15e36ff505558842da29023f3f450441b771e3244bdc0", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/mx_creative_writing_creative_writing/al_curve.csv", "source_sha256": "0c193663d6bf65c6ffede829949cb80bc68801c4b52e7b3b254c410cb476e8d3", "staged_path": "results/historical-al-curves/v1/mx_creative_writing_creative_writing/al_curve.csv", "staged_sha256": "0c193663d6bf65c6ffede829949cb80bc68801c4b52e7b3b254c410cb476e8d3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 93, "source_path": "experiments/dflash/eval_csv/mx_creative_writing_factual_qa/al_curve.csv", "source_sha256": "169dcde892e3f47236c2e5a440b18478aac40b6d578920e865124e3ce1ce729e", "staged_path": "results/historical-al-curves/v1/mx_creative_writing_factual_qa/al_curve.csv", "staged_sha256": "169dcde892e3f47236c2e5a440b18478aac40b6d578920e865124e3ce1ce729e", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 88, "source_path": "experiments/dflash/eval_csv/mx_creative_writing_general/al_curve.csv", "source_sha256": "79ffe500ee72208d201cde35dada407cd3120dd3761ebb53cc89107efac2420b", "staged_path": "results/historical-al-curves/v1/mx_creative_writing_general/al_curve.csv", "staged_sha256": "79ffe500ee72208d201cde35dada407cd3120dd3761ebb53cc89107efac2420b", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 85, "source_path": "experiments/dflash/eval_csv/mx_creative_writing_math/al_curve.csv", "source_sha256": "4cb4168637b1d2ecfc71fe0709a3cabc8a59dbdf0d4a8894af7473afb3fe5458", "staged_path": "results/historical-al-curves/v1/mx_creative_writing_math/al_curve.csv", "staged_sha256": "4cb4168637b1d2ecfc71fe0709a3cabc8a59dbdf0d4a8894af7473afb3fe5458", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/mx_factual_qa_code/al_curve.csv", "source_sha256": "0da911fc412c8831a08db40720644b2d07d1878e8b2d9505d2c344c58ed18f32", "staged_path": "results/historical-al-curves/v1/mx_factual_qa_code/al_curve.csv", "staged_sha256": "0da911fc412c8831a08db40720644b2d07d1878e8b2d9505d2c344c58ed18f32", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 93, "source_path": "experiments/dflash/eval_csv/mx_factual_qa_creative_writing/al_curve.csv", "source_sha256": "c22215e9aac307a3d72ec1dc4054e8c65bbfaae5e81c501d2e628b89e0ccac6a", "staged_path": "results/historical-al-curves/v1/mx_factual_qa_creative_writing/al_curve.csv", "staged_sha256": "c22215e9aac307a3d72ec1dc4054e8c65bbfaae5e81c501d2e628b89e0ccac6a", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/mx_factual_qa_factual_qa/al_curve.csv", "source_sha256": "ebc0bf6bf21573185e2a4e494f4d2ba3a3fdefc0ab5bdcf7dd6cbb77bdb63e14", "staged_path": "results/historical-al-curves/v1/mx_factual_qa_factual_qa/al_curve.csv", "staged_sha256": "ebc0bf6bf21573185e2a4e494f4d2ba3a3fdefc0ab5bdcf7dd6cbb77bdb63e14", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 82, "source_path": "experiments/dflash/eval_csv/mx_factual_qa_general/al_curve.csv", "source_sha256": "42edb75d5a4adece22f31fc5895568f5a0195373097fd356b1f463afaf03fca7", "staged_path": "results/historical-al-curves/v1/mx_factual_qa_general/al_curve.csv", "staged_sha256": "42edb75d5a4adece22f31fc5895568f5a0195373097fd356b1f463afaf03fca7", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/mx_factual_qa_math/al_curve.csv", "source_sha256": "29dfae8c16ec355149bec331dab105867ec8741f0af6dc411d15d9074f2927ce", "staged_path": "results/historical-al-curves/v1/mx_factual_qa_math/al_curve.csv", "staged_sha256": "29dfae8c16ec355149bec331dab105867ec8741f0af6dc411d15d9074f2927ce", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 76, "source_path": "experiments/dflash/eval_csv/mx_general_code/al_curve.csv", "source_sha256": "ea26ed3625839230a90b0b5fb91ad8808856c13e16e1e303cf799c736c5b570d", "staged_path": "results/historical-al-curves/v1/mx_general_code/al_curve.csv", "staged_sha256": "ea26ed3625839230a90b0b5fb91ad8808856c13e16e1e303cf799c736c5b570d", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 90, "source_path": "experiments/dflash/eval_csv/mx_general_creative_writing/al_curve.csv", "source_sha256": "8e0fb5a0a6730b5f416190e30515cf294b4646c09d285f070d8d3eb15e0ce6bf", "staged_path": "results/historical-al-curves/v1/mx_general_creative_writing/al_curve.csv", "staged_sha256": "8e0fb5a0a6730b5f416190e30515cf294b4646c09d285f070d8d3eb15e0ce6bf", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 83, "source_path": "experiments/dflash/eval_csv/mx_general_factual_qa/al_curve.csv", "source_sha256": "d8767ce37119ecb19671fe82e28c340b6323d8c5fcce7e336ee004cc30472499", "staged_path": "results/historical-al-curves/v1/mx_general_factual_qa/al_curve.csv", "staged_sha256": "d8767ce37119ecb19671fe82e28c340b6323d8c5fcce7e336ee004cc30472499", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 80, "source_path": "experiments/dflash/eval_csv/mx_general_general/al_curve.csv", "source_sha256": "127bbd05b322c4e702ca8637c1754d299986f0ef8d08a9a37f8d734bb22e7ead", "staged_path": "results/historical-al-curves/v1/mx_general_general/al_curve.csv", "staged_sha256": "127bbd05b322c4e702ca8637c1754d299986f0ef8d08a9a37f8d734bb22e7ead", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 77, "source_path": "experiments/dflash/eval_csv/mx_general_math/al_curve.csv", "source_sha256": "3701966065391c5c89b1dca082fd9ab93ebc4ee444b51db0ae8d05268a5a50f1", "staged_path": "results/historical-al-curves/v1/mx_general_math/al_curve.csv", "staged_sha256": "3701966065391c5c89b1dca082fd9ab93ebc4ee444b51db0ae8d05268a5a50f1", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 75, "source_path": "experiments/dflash/eval_csv/mx_math_code/al_curve.csv", "source_sha256": "a6b1afa6900cb541c4979a6714d42aed473565d9d646088961e2f5c4ccb6736b", "staged_path": "results/historical-al-curves/v1/mx_math_code/al_curve.csv", "staged_sha256": "a6b1afa6900cb541c4979a6714d42aed473565d9d646088961e2f5c4ccb6736b", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 87, "source_path": "experiments/dflash/eval_csv/mx_math_creative_writing/al_curve.csv", "source_sha256": "5eba966fd12aaf38b508164c325bb480d95aceec7a8fd3b359ff12084c8b8635", "staged_path": "results/historical-al-curves/v1/mx_math_creative_writing/al_curve.csv", "staged_sha256": "5eba966fd12aaf38b508164c325bb480d95aceec7a8fd3b359ff12084c8b8635", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 81, "source_path": "experiments/dflash/eval_csv/mx_math_factual_qa/al_curve.csv", "source_sha256": "42c6c251cfe5c3559195b53a9860f00e545286f24aacf56e700f5d8c24c29aa8", "staged_path": "results/historical-al-curves/v1/mx_math_factual_qa/al_curve.csv", "staged_sha256": "42c6c251cfe5c3559195b53a9860f00e545286f24aacf56e700f5d8c24c29aa8", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 78, "source_path": "experiments/dflash/eval_csv/mx_math_general/al_curve.csv", "source_sha256": "0ef053311d4cae8ba0d339bf066e396bbc9624fccb00ee79a9ccf6215190f772", "staged_path": "results/historical-al-curves/v1/mx_math_general/al_curve.csv", "staged_sha256": "0ef053311d4cae8ba0d339bf066e396bbc9624fccb00ee79a9ccf6215190f772", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 73, "source_path": "experiments/dflash/eval_csv/mx_math_math/al_curve.csv", "source_sha256": "601ddb84962c2e9a12c49338d3688b534b1b6e8efc3ea1f7c47dfd8274ac3e85", "staged_path": "results/historical-al-curves/v1/mx_math_math/al_curve.csv", "staged_sha256": "601ddb84962c2e9a12c49338d3688b534b1b6e8efc3ea1f7c47dfd8274ac3e85", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 458, "source_path": "experiments/dflash/eval_csv/rc89/al_curve.csv", "source_sha256": "ae5e4086fd51850db624e71aad1f6390564afea4c01909138b8ec518174a6ddd", "staged_path": "results/historical-al-curves/v1/rc89/al_curve.csv", "staged_sha256": "ae5e4086fd51850db624e71aad1f6390564afea4c01909138b8ec518174a6ddd", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 288, "source_path": "experiments/dflash/eval_csv/route89/al_curve.csv", "source_sha256": "1e682fc087dfa045ab6cd0604b6a6d60ccc9779188c383b3cce5d98922cb3348", "staged_path": "results/historical-al-curves/v1/route89/al_curve.csv", "staged_sha256": "1e682fc087dfa045ab6cd0604b6a6d60ccc9779188c383b3cce5d98922cb3348", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 222, "source_path": "experiments/dflash/eval_csv/rw89x/al_curve.csv", "source_sha256": "226d571094d6326d5c77156c8b6f9b6f23565eb0a558589f23bc047a89b59fb9", "staged_path": "results/historical-al-curves/v1/rw89x/al_curve.csv", "staged_sha256": "226d571094d6326d5c77156c8b6f9b6f23565eb0a558589f23bc047a89b59fb9", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/rwr2_0_code/al_curve.csv", "source_sha256": "d3786c58dd808b47532eadae27684b67369fd668725e3c58ab5b6d95fbb8eb0f", "staged_path": "results/historical-al-curves/v1/rwr2_0_code/al_curve.csv", "staged_sha256": "d3786c58dd808b47532eadae27684b67369fd668725e3c58ab5b6d95fbb8eb0f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 101, "source_path": "experiments/dflash/eval_csv/rwr2_1_code/al_curve.csv", "source_sha256": "98daad9455c1beef6de932d4c515c3f211ecce2b6db33d4195d2882f2a4ec8a0", "staged_path": "results/historical-al-curves/v1/rwr2_1_code/al_curve.csv", "staged_sha256": "98daad9455c1beef6de932d4c515c3f211ecce2b6db33d4195d2882f2a4ec8a0", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 101, "source_path": "experiments/dflash/eval_csv/rwr2_2_code/al_curve.csv", "source_sha256": "ee1dc53951abdc88505e4eb04a160a2a0c069529ae4c7921f4a161c087bc7ed7", "staged_path": "results/historical-al-curves/v1/rwr2_2_code/al_curve.csv", "staged_sha256": "ee1dc53951abdc88505e4eb04a160a2a0c069529ae4c7921f4a161c087bc7ed7", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 93, "source_path": "experiments/dflash/eval_csv/scnAg0_code/al_curve.csv", "source_sha256": "d66a15fbbacb240934dc919517d50be294ee76085c831d44adfc4131b9f9a324", "staged_path": "results/historical-al-curves/v1/scnAg0_code/al_curve.csv", "staged_sha256": "d66a15fbbacb240934dc919517d50be294ee76085c831d44adfc4131b9f9a324", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 93, "source_path": "experiments/dflash/eval_csv/scnAg0_math/al_curve.csv", "source_sha256": "c0d060695f2f0c87fcd74cd065b720d47548eaf06d91dbc942fd1a8aa5d179ca", "staged_path": "results/historical-al-curves/v1/scnAg0_math/al_curve.csv", "staged_sha256": "c0d060695f2f0c87fcd74cd065b720d47548eaf06d91dbc942fd1a8aa5d179ca", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 94, "source_path": "experiments/dflash/eval_csv/scnAg1_code/al_curve.csv", "source_sha256": "071f37849a80671bebc47c9c9bd9020b0a150dc39185575f323621cf5808e6a2", "staged_path": "results/historical-al-curves/v1/scnAg1_code/al_curve.csv", "staged_sha256": "071f37849a80671bebc47c9c9bd9020b0a150dc39185575f323621cf5808e6a2", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 79, "source_path": "experiments/dflash/eval_csv/scnAg1_math/al_curve.csv", "source_sha256": "f74a58358baddb9ad7038e3be19273d990e93fd1f6814d2c14b2fe9fe2a840ab", "staged_path": "results/historical-al-curves/v1/scnAg1_math/al_curve.csv", "staged_sha256": "f74a58358baddb9ad7038e3be19273d990e93fd1f6814d2c14b2fe9fe2a840ab", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 92, "source_path": "experiments/dflash/eval_csv/scnAg2_code/al_curve.csv", "source_sha256": "e81f887a2721c5e0334c956f7a4de7f31102b5a76b453e8510c34883172148d3", "staged_path": "results/historical-al-curves/v1/scnAg2_code/al_curve.csv", "staged_sha256": "e81f887a2721c5e0334c956f7a4de7f31102b5a76b453e8510c34883172148d3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 93, "source_path": "experiments/dflash/eval_csv/scnAg2_math/al_curve.csv", "source_sha256": "b8d5de6fafb9437dd34f8dd0a92b861e3ff4303e21d37fdd82ea0ba426563025", "staged_path": "results/historical-al-curves/v1/scnAg2_math/al_curve.csv", "staged_sha256": "b8d5de6fafb9437dd34f8dd0a92b861e3ff4303e21d37fdd82ea0ba426563025", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/scnAg3_code/al_curve.csv", "source_sha256": "d177d4bd11c69bb5a669851dd02451a604ebeceb9d9763594d5f9f083bf75c85", "staged_path": "results/historical-al-curves/v1/scnAg3_code/al_curve.csv", "staged_sha256": "d177d4bd11c69bb5a669851dd02451a604ebeceb9d9763594d5f9f083bf75c85", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 96, "source_path": "experiments/dflash/eval_csv/scnAg3_math/al_curve.csv", "source_sha256": "81a5d982e68efcd144f0353c18f97d1eb5372018d1e7bcdd0ebff652b15e16f7", "staged_path": "results/historical-al-curves/v1/scnAg3_math/al_curve.csv", "staged_sha256": "81a5d982e68efcd144f0353c18f97d1eb5372018d1e7bcdd0ebff652b15e16f7", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 93, "source_path": "experiments/dflash/eval_csv/scnAm0_math/al_curve.csv", "source_sha256": "7de24f2594949bfe5ee6faa5ca99d976d72e97741ea961315aebfe1e447bd0d8", "staged_path": "results/historical-al-curves/v1/scnAm0_math/al_curve.csv", "staged_sha256": "7de24f2594949bfe5ee6faa5ca99d976d72e97741ea961315aebfe1e447bd0d8", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 93, "source_path": "experiments/dflash/eval_csv/scnAm1_math/al_curve.csv", "source_sha256": "3cbab716c23dd154accda87f4bcf794b2fc0651752aad17d823dc39b062b1660", "staged_path": "results/historical-al-curves/v1/scnAm1_math/al_curve.csv", "staged_sha256": "3cbab716c23dd154accda87f4bcf794b2fc0651752aad17d823dc39b062b1660", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 93, "source_path": "experiments/dflash/eval_csv/scnAm2_math/al_curve.csv", "source_sha256": "bf5646e0311035d6a58c3aa05f4c16c5a03ec9cad2a5e3f238d0870a468c932f", "staged_path": "results/historical-al-curves/v1/scnAm2_math/al_curve.csv", "staged_sha256": "bf5646e0311035d6a58c3aa05f4c16c5a03ec9cad2a5e3f238d0870a468c932f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/scnAm3_math/al_curve.csv", "source_sha256": "d1d8c2c472413573c1649ed6ece2cf45351f0f2801b90d3c1c3c1765de330ad0", "staged_path": "results/historical-al-curves/v1/scnAm3_math/al_curve.csv", "staged_sha256": "d1d8c2c472413573c1649ed6ece2cf45351f0f2801b90d3c1c3c1765de330ad0", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/scnAm_code/al_curve.csv", "source_sha256": "a27520a897853df3e676180297494c22b731734ac6b01dd7128ab0e28744dcae", "staged_path": "results/historical-al-curves/v1/scnAm_code/al_curve.csv", "staged_sha256": "a27520a897853df3e676180297494c22b731734ac6b01dd7128ab0e28744dcae", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 93, "source_path": "experiments/dflash/eval_csv/sd0_code/al_curve.csv", "source_sha256": "df9636a07e3c7c635e4c25e7c4c2ec39a062c3b691ed01d7630c902de36665cf", "staged_path": "results/historical-al-curves/v1/sd0_code/al_curve.csv", "staged_sha256": "df9636a07e3c7c635e4c25e7c4c2ec39a062c3b691ed01d7630c902de36665cf", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/sd1_code/al_curve.csv", "source_sha256": "50f86a08650e19fdec4c1705af72c2667a25ad12c21fd228f540dfc17807fd91", "staged_path": "results/historical-al-curves/v1/sd1_code/al_curve.csv", "staged_sha256": "50f86a08650e19fdec4c1705af72c2667a25ad12c21fd228f540dfc17807fd91", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/sd2_code/al_curve.csv", "source_sha256": "b966d37e18be328c18ac4488c53ca7ad314680d6878ba721b1f0b126bfde4351", "staged_path": "results/historical-al-curves/v1/sd2_code/al_curve.csv", "staged_sha256": "b966d37e18be328c18ac4488c53ca7ad314680d6878ba721b1f0b126bfde4351", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/sd3_code/al_curve.csv", "source_sha256": "595ca5c6d0f9c564c9c4d3c1858259f14ba43a6c59a2c3a980017c8dddea6d5c", "staged_path": "results/historical-al-curves/v1/sd3_code/al_curve.csv", "staged_sha256": "595ca5c6d0f9c564c9c4d3c1858259f14ba43a6c59a2c3a980017c8dddea6d5c", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 94, "source_path": "experiments/dflash/eval_csv/sd4_code/al_curve.csv", "source_sha256": "49f06e20a64540e35a1882f0d6f77c2ef97b0de40221adbdd4ce5ebb2e4fa6cc", "staged_path": "results/historical-al-curves/v1/sd4_code/al_curve.csv", "staged_sha256": "49f06e20a64540e35a1882f0d6f77c2ef97b0de40221adbdd4ce5ebb2e4fa6cc", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/sd5_code/al_curve.csv", "source_sha256": "203fe4ec71adef91a5e2c932d23b23a55f16b4db72cf8c8c26c10391807247f3", "staged_path": "results/historical-al-curves/v1/sd5_code/al_curve.csv", "staged_sha256": "203fe4ec71adef91a5e2c932d23b23a55f16b4db72cf8c8c26c10391807247f3", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 108, "source_path": "experiments/dflash/eval_csv/sd5pk_creative_writing/al_curve.csv", "source_sha256": "93e9532b8afa55c7d04ebf21a097f5050719ae7faaf79c514a0b49fdb76433dd", "staged_path": "results/historical-al-curves/v1/sd5pk_creative_writing/al_curve.csv", "staged_sha256": "93e9532b8afa55c7d04ebf21a097f5050719ae7faaf79c514a0b49fdb76433dd", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 101, "source_path": "experiments/dflash/eval_csv/sd5pk_factual_qa/al_curve.csv", "source_sha256": "6b48f906b7b3fc0ec47d1b4cb5a758c6779bae518405ab96773a31add496885e", "staged_path": "results/historical-al-curves/v1/sd5pk_factual_qa/al_curve.csv", "staged_sha256": "6b48f906b7b3fc0ec47d1b4cb5a758c6779bae518405ab96773a31add496885e", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 98, "source_path": "experiments/dflash/eval_csv/sd5pk_general/al_curve.csv", "source_sha256": "b024f11ac64c058c715cb5652556058e912dc5b2b27e6a45f1c2a98ff9839b20", "staged_path": "results/historical-al-curves/v1/sd5pk_general/al_curve.csv", "staged_sha256": "b024f11ac64c058c715cb5652556058e912dc5b2b27e6a45f1c2a98ff9839b20", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 96, "source_path": "experiments/dflash/eval_csv/sd5pk_math/al_curve.csv", "source_sha256": "86a5dc6aa38a11753c1f7c543bac0be7e9508412fd9d3621b8a1dccf675a52b7", "staged_path": "results/historical-al-curves/v1/sd5pk_math/al_curve.csv", "staged_sha256": "86a5dc6aa38a11753c1f7c543bac0be7e9508412fd9d3621b8a1dccf675a52b7", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/sd6_code/al_curve.csv", "source_sha256": "bd8c6ca5ee8bbc993cceb689528c6324f0c06175684ade3addd753d4e30bd124", "staged_path": "results/historical-al-curves/v1/sd6_code/al_curve.csv", "staged_sha256": "bd8c6ca5ee8bbc993cceb689528c6324f0c06175684ade3addd753d4e30bd124", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 96, "source_path": "experiments/dflash/eval_csv/sd7_code/al_curve.csv", "source_sha256": "60e6b8260e89c9efab753e867c2bde1b4a8ee54b1715a7215216849297f33fb8", "staged_path": "results/historical-al-curves/v1/sd7_code/al_curve.csv", "staged_sha256": "60e6b8260e89c9efab753e867c2bde1b4a8ee54b1715a7215216849297f33fb8", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/sg0_code/al_curve.csv", "source_sha256": "e803f1bd631885a062f8cf78cfc78fd6010d5a75dd7a6575fc71256d386a5d91", "staged_path": "results/historical-al-curves/v1/sg0_code/al_curve.csv", "staged_sha256": "e803f1bd631885a062f8cf78cfc78fd6010d5a75dd7a6575fc71256d386a5d91", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 96, "source_path": "experiments/dflash/eval_csv/sg1_code/al_curve.csv", "source_sha256": "99e0a6ef3217f4d18ddc84092678467918f548c52e2fac892f779dade359cad5", "staged_path": "results/historical-al-curves/v1/sg1_code/al_curve.csv", "staged_sha256": "99e0a6ef3217f4d18ddc84092678467918f548c52e2fac892f779dade359cad5", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 96, "source_path": "experiments/dflash/eval_csv/sg2_code/al_curve.csv", "source_sha256": "2312de5495b1bb06fd039a8ca49124b2ea19ae82174889c996aba3af5890cba7", "staged_path": "results/historical-al-curves/v1/sg2_code/al_curve.csv", "staged_sha256": "2312de5495b1bb06fd039a8ca49124b2ea19ae82174889c996aba3af5890cba7", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 97, "source_path": "experiments/dflash/eval_csv/sg3_code/al_curve.csv", "source_sha256": "3abbfc13e520c00535e6cd2ecff9aa5aa91b8e3e4f1feca133a8c4557c580dbb", "staged_path": "results/historical-al-curves/v1/sg3_code/al_curve.csv", "staged_sha256": "3abbfc13e520c00535e6cd2ecff9aa5aa91b8e3e4f1feca133a8c4557c580dbb", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 95, "source_path": "experiments/dflash/eval_csv/sg4_code/al_curve.csv", "source_sha256": "86a4d3bdeca1c89716a4afc24dc1f9a56a14e8df479fc8071fca91f8b0b3da05", "staged_path": "results/historical-al-curves/v1/sg4_code/al_curve.csv", "staged_sha256": "86a4d3bdeca1c89716a4afc24dc1f9a56a14e8df479fc8071fca91f8b0b3da05", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 97, "source_path": "experiments/dflash/eval_csv/sg5_code/al_curve.csv", "source_sha256": "e5b0bec024da1370030d6c903dc14a353aa22d92a6c1a1efe2116374307265a8", "staged_path": "results/historical-al-curves/v1/sg5_code/al_curve.csv", "staged_sha256": "e5b0bec024da1370030d6c903dc14a353aa22d92a6c1a1efe2116374307265a8", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 97, "source_path": "experiments/dflash/eval_csv/sg6_code/al_curve.csv", "source_sha256": "182709212d475a24728d84a642a0f589d0ad3807183a40a5375a95998f9ab05b", "staged_path": "results/historical-al-curves/v1/sg6_code/al_curve.csv", "staged_sha256": "182709212d475a24728d84a642a0f589d0ad3807183a40a5375a95998f9ab05b", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 97, "source_path": "experiments/dflash/eval_csv/sg7_code/al_curve.csv", "source_sha256": "171d83662652e87ba1819c8e3ed696d30ae974994d486f50904a8673155c15a4", "staged_path": "results/historical-al-curves/v1/sg7_code/al_curve.csv", "staged_sha256": "171d83662652e87ba1819c8e3ed696d30ae974994d486f50904a8673155c15a4", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 109, "source_path": "experiments/dflash/eval_csv/sgpk_creative_writing/al_curve.csv", "source_sha256": "763cb159181b98cffc15de07014e37bf2c378c84953a02147d860bfb76ff9f64", "staged_path": "results/historical-al-curves/v1/sgpk_creative_writing/al_curve.csv", "staged_sha256": "763cb159181b98cffc15de07014e37bf2c378c84953a02147d860bfb76ff9f64", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 103, "source_path": "experiments/dflash/eval_csv/sgpk_factual_qa/al_curve.csv", "source_sha256": "94d277ef87c5eb724a44e1e4e6e4bd310c2e86415ec10069c4cd65d111baf549", "staged_path": "results/historical-al-curves/v1/sgpk_factual_qa/al_curve.csv", "staged_sha256": "94d277ef87c5eb724a44e1e4e6e4bd310c2e86415ec10069c4cd65d111baf549", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 99, "source_path": "experiments/dflash/eval_csv/sgpk_general/al_curve.csv", "source_sha256": "d9d8404db125c2cf0594ec8f30cc2f95fdf144eb2768dd1518dad1ad251bd08f", "staged_path": "results/historical-al-curves/v1/sgpk_general/al_curve.csv", "staged_sha256": "d9d8404db125c2cf0594ec8f30cc2f95fdf144eb2768dd1518dad1ad251bd08f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 96, "source_path": "experiments/dflash/eval_csv/sgpk_math/al_curve.csv", "source_sha256": "566138ead9cd5d33e7434ac677ef7064c85c90587017164f9b12c89942b9f2d5", "staged_path": "results/historical-al-curves/v1/sgpk_math/al_curve.csv", "staged_sha256": "566138ead9cd5d33e7434ac677ef7064c85c90587017164f9b12c89942b9f2d5", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 107, "source_path": "experiments/dflash/eval_csv/wr2be0_code/al_curve.csv", "source_sha256": "b21b54dcc70720f4b52af10e8d85653072580c197b8020c128871e208a462f9f", "staged_path": "results/historical-al-curves/v1/wr2be0_code/al_curve.csv", "staged_sha256": "b21b54dcc70720f4b52af10e8d85653072580c197b8020c128871e208a462f9f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 107, "source_path": "experiments/dflash/eval_csv/wr2be1_code/al_curve.csv", "source_sha256": "79a52b7a1a9ad1aa7412ec94606f1e762c1b5ede13fa3ea5c2a6d6da6509200f", "staged_path": "results/historical-al-curves/v1/wr2be1_code/al_curve.csv", "staged_sha256": "79a52b7a1a9ad1aa7412ec94606f1e762c1b5ede13fa3ea5c2a6d6da6509200f", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 106, "source_path": "experiments/dflash/eval_csv/wr2be2_code/al_curve.csv", "source_sha256": "b8c3d0f74241fd61dc13cb394b7780c546160e26730a0956d1938285f096a65e", "staged_path": "results/historical-al-curves/v1/wr2be2_code/al_curve.csv", "staged_sha256": "b8c3d0f74241fd61dc13cb394b7780c546160e26730a0956d1938285f096a65e", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 102, "source_path": "experiments/dflash/eval_csv/wr3e0_code/al_curve.csv", "source_sha256": "fe39d6da277ea0f5fcaa73778bb46c2ef01a8309487153963f92764d1dc8d780", "staged_path": "results/historical-al-curves/v1/wr3e0_code/al_curve.csv", "staged_sha256": "fe39d6da277ea0f5fcaa73778bb46c2ef01a8309487153963f92764d1dc8d780", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 101, "source_path": "experiments/dflash/eval_csv/wr3e1_code/al_curve.csv", "source_sha256": "dec31dbef3209b9019a41fe89f9c6a586f4a3dc0878287df741953101288a583", "staged_path": "results/historical-al-curves/v1/wr3e1_code/al_curve.csv", "staged_sha256": "dec31dbef3209b9019a41fe89f9c6a586f4a3dc0878287df741953101288a583", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-al-curve", "license": "mit", "size_bytes": 102, "source_path": "experiments/dflash/eval_csv/wr3e2_code/al_curve.csv", "source_sha256": "6cfca51dd583996e8e68b1f2659aa472c2445ab4e0a847ef4d6a796276ae16fd", "staged_path": "results/historical-al-curves/v1/wr3e2_code/al_curve.csv", "staged_sha256": "6cfca51dd583996e8e68b1f2659aa472c2445ab4e0a847ef4d6a796276ae16fd", "status": "historical", "transformation": "direct-copy"} +{"artifact_type": "aggregate-main-figure-evidence", "license": "mit", "size_bytes": 2695, "source_path": "paper/submission/evidence/r1_mainfig_seed20260719_20260721T0602Z_cells_summary.json", "source_sha256": "358813a37e84a8e97e1d16299277f0b9e80c39a1d9230e6eddb1d128afa153cb", "staged_path": "results/main-figure-r1/v1/frozen-aggregate-cells.json", "staged_sha256": "7e858ff1246214c81cf70976c920436d3271f9c77bd4e13a659ec9f4d3768864", "status": "frozen-aggregate-52-of-52", "transformation": "sanitized-json-projection; retained aggregate panels and completion metadata only; removed cells, run_root, sidecars, prompt references, source checkpoints, and file paths"} +{"artifact_type": "verification-recipe", "license": "mit", "size_bytes": 17119, "source_path": "experiments/dflash/scripts/bootstrap_b1_exact_3p2m_routed.py", "source_sha256": "b2b335f41aa9a2381726c9d2c90aa8eb12a1a3b4f31ebfa60c31045ec168855d", "staged_path": "verification/v1/bootstrap_b1_exact_3p2m_routed.py", "staged_sha256": "b2b335f41aa9a2381726c9d2c90aa8eb12a1a3b4f31ebfa60c31045ec168855d", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "verification-recipe", "license": "mit", "size_bytes": 16381, "source_path": "experiments/dflash/scripts/bootstrap_b2_epoch5_router.py", "source_sha256": "aee31b4fac56dba2507403e989daf7010a50a01fde3edb7ebca0bad57a90f2ae", "staged_path": "verification/v1/bootstrap_b2_epoch5_router.py", "staged_sha256": "aee31b4fac56dba2507403e989daf7010a50a01fde3edb7ebca0bad57a90f2ae", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "verification-recipe", "license": "mit", "size_bytes": 31731, "source_path": "experiments/dflash/scripts/bootstrap_paired_al.py", "source_sha256": "a20de477a77313a08619b1977b80e7fb7ffe83bd08bd923e0aa5c8132dd6901d", "staged_path": "verification/v1/bootstrap_paired_al.py", "staged_sha256": "a20de477a77313a08619b1977b80e7fb7ffe83bd08bd923e0aa5c8132dd6901d", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "verification-recipe", "license": "mit", "size_bytes": 6819, "source_path": "experiments/dflash/scripts/r1_mainfig_verify_exports.py", "source_sha256": "9734fc9b2b24a74ffa63708ae0ae7442962cef40f9147c14c3528c416c78d0ca", "staged_path": "verification/v1/r1_mainfig_verify_exports.py", "staged_sha256": "9734fc9b2b24a74ffa63708ae0ae7442962cef40f9147c14c3528c416c78d0ca", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "verification-recipe", "license": "mit", "size_bytes": 10205, "source_path": "experiments/dflash/scripts/verify_b1_exact_3p2m_routed_artifacts.py", "source_sha256": "f52bef510022d3ca470104bb184cdffe83ead49648dd712e64b0ea9149d901e3", "staged_path": "verification/v1/verify_b1_exact_3p2m_routed_artifacts.py", "staged_sha256": "f52bef510022d3ca470104bb184cdffe83ead49648dd712e64b0ea9149d901e3", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "verification-recipe", "license": "mit", "size_bytes": 17305, "source_path": "experiments/dflash/scripts/verify_b2_epoch5_artifacts.py", "source_sha256": "7427838713af5748fbfd600af2b8a8d4d50d58a7403f072c352c025b12a0e3d1", "staged_path": "verification/v1/verify_b2_epoch5_artifacts.py", "staged_sha256": "7427838713af5748fbfd600af2b8a8d4d50d58a7403f072c352c025b12a0e3d1", "status": "reusable-code", "transformation": "direct-copy"} +{"artifact_type": "verification-recipe", "license": "mit", "size_bytes": 4229, "source_path": "experiments/dflash/scripts/verify_b5_qwen3_4b_evidence.py", "source_sha256": "67e2df392b3f69b5e9e8d4c0c2dddc20f7e5b79a9dcbb7f8cd3fffd278cf5a33", "staged_path": "verification/v1/verify_b5_qwen3_4b_evidence.py", "staged_sha256": "67e2df392b3f69b5e9e8d4c0c2dddc20f7e5b79a9dcbb7f8cd3fffd278cf5a33", "status": "reusable-code", "transformation": "direct-copy"} diff --git a/manifests/extended-evidence-v1/release-summary.json b/manifests/extended-evidence-v1/release-summary.json new file mode 100644 index 0000000000000000000000000000000000000000..8d86f4c01b0bf7592533b3fc5be84180cd01f8ca --- /dev/null +++ b/manifests/extended-evidence-v1/release-summary.json @@ -0,0 +1,34 @@ +{ + "artifact_type_counts": { + "aggregate-al-curve": 249, + "aggregate-main-figure-evidence": 1, + "documentation": 2, + "environment-recipe": 1, + "figure-data": 3, + "license": 1, + "plotting-recipe": 15, + "rendered-figure": 22, + "rendered-figure-or-source": 6, + "verification-recipe": 7 + }, + "bytes": 5833027, + "files": 307, + "freeze_id": "extended-evidence-v1", + "freeze_time": "2026-07-23T20:26:58Z", + "license": "mit", + "safety_scan": { + "absolute_internal_paths": "pass", + "credentials": "pass", + "private_internal_material": "absent", + "raw_logs": "absent", + "raw_prompts_or_generations": "absent" + }, + "transformations": { + "direct-copy": 293, + "generated-from-selected-script-imports": 1, + "generated-public-documentation": 2, + "sanitized-copy; replaced 1 hard-coded output path(s) with relative output path(s)": 9, + "sanitized-copy; replaced 2 hard-coded output path(s) with relative output path(s)": 1, + "sanitized-json-projection; retained aggregate panels and completion metadata only; removed cells, run_root, sidecars, prompt references, source checkpoints, and file paths": 1 + } +} diff --git a/recipes/plotting/v1/build_mos_architecture_drawio.py b/recipes/plotting/v1/build_mos_architecture_drawio.py new file mode 100644 index 0000000000000000000000000000000000000000..4d34ee998398ab7f7a0ea39f1371c37f5b834623 --- /dev/null +++ b/recipes/plotting/v1/build_mos_architecture_drawio.py @@ -0,0 +1,856 @@ +#!/usr/bin/env python3 +"""Build a clean, editable draw.io source for the MoS architecture figure. + +The script only writes draw.io XML. Export and visual verification are run on +the server so the local desktop environment never invokes the draw.io CLI. +""" + +from __future__ import annotations + +import argparse +from pathlib import Path +import xml.etree.ElementTree as ET + + +W, H = 2400, 1420 + +INK = "#263746" +MUTED = "#6D7C87" +RULE = "#D7E0E5" +SURFACE = "#FBFCFD" +WHITE = "#FFFFFF" + +SHARED = "#5F879C" +SHARED_DARK = "#355F74" +SHARED_FILL = "#EAF2F6" + +MLP = "#7A62A0" +MLP_DARK = "#5C477B" +MLP_FILL = "#EEEAF5" +MLP_FAINT = "#F7F5FA" + +ROUTE = "#4F8E83" +ROUTE_DARK = "#2B6D64" +ROUTE_FILL = "#E9F3F1" + +FROZEN = "#9AA7AF" +FROZEN_DARK = "#6D7B84" +FROZEN_FILL = "#F1F4F5" + +GOOD = "#4F9169" +GOOD_FILL = "#EAF4ED" +AMBER = "#C77E26" +AMBER_FILL = "#FFF1DD" + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--output", type=Path, required=True) + return parser.parse_args() + + +class Diagram: + def __init__(self) -> None: + self.mxfile = ET.Element( + "mxfile", + {"host": "Electron", "modified": "2026-07-21T00:00:00.000Z", "version": "30.3.14"}, + ) + page = ET.SubElement(self.mxfile, "diagram", {"id": "mos-v8", "name": "MoS architecture"}) + self.model = ET.SubElement( + page, + "mxGraphModel", + { + "dx": "0", + "dy": "0", + "grid": "1", + "gridSize": "10", + "guides": "1", + "tooltips": "1", + "connect": "1", + "arrows": "1", + "fold": "1", + "page": "1", + "pageScale": "1", + "pageWidth": str(W), + "pageHeight": str(H), + "math": "0", + "shadow": "0", + "background": WHITE, + }, + ) + self.root = ET.SubElement(self.model, "root") + ET.SubElement(self.root, "mxCell", {"id": "0"}) + ET.SubElement(self.root, "mxCell", {"id": "1", "parent": "0"}) + self.counter = 2 + + def new_id(self, prefix: str) -> str: + cell_id = f"{prefix}_{self.counter}" + self.counter += 1 + return cell_id + + def vertex( + self, + value: str, + x: int, + y: int, + w: int, + h: int, + style: str, + *, + cell_id: str | None = None, + parent: str = "1", + ) -> str: + cell_id = cell_id or self.new_id("v") + cell = ET.SubElement( + self.root, + "mxCell", + {"id": cell_id, "value": value, "style": style, "vertex": "1", "parent": parent}, + ) + ET.SubElement( + cell, + "mxGeometry", + {"x": str(x), "y": str(y), "width": str(w), "height": str(h), "as": "geometry"}, + ) + return cell_id + + def edge( + self, + source: str, + target: str, + *, + color: str = INK, + width: float = 2.0, + dashed: bool = False, + exit_xy: tuple[float, float] | None = None, + entry_xy: tuple[float, float] | None = None, + points: list[tuple[int, int]] | None = None, + end_arrow: str = "blockThin", + cell_id: str | None = None, + ) -> str: + cell_id = cell_id or self.new_id("e") + style = [ + "edgeStyle=orthogonalEdgeStyle", + "rounded=1", + "orthogonalLoop=1", + "jettySize=auto", + "html=1", + "convertToSvg=1", + f"strokeColor={color}", + f"strokeWidth={width}", + f"endArrow={end_arrow}", + "endFill=1" if end_arrow != "none" else "endFill=0", + "endSize=9", + ] + if dashed: + style.extend(["dashed=1", "dashPattern=8 6"]) + if exit_xy is not None: + style.extend([f"exitX={exit_xy[0]}", f"exitY={exit_xy[1]}", "exitDx=0", "exitDy=0"]) + if entry_xy is not None: + style.extend([f"entryX={entry_xy[0]}", f"entryY={entry_xy[1]}", "entryDx=0", "entryDy=0"]) + cell = ET.SubElement( + self.root, + "mxCell", + { + "id": cell_id, + "value": "", + "style": ";".join(style) + ";", + "edge": "1", + "parent": "1", + "source": source, + "target": target, + }, + ) + geom = ET.SubElement(cell, "mxGeometry", {"relative": "1", "as": "geometry"}) + if points: + array = ET.SubElement(geom, "Array", {"as": "points"}) + for px, py in points: + ET.SubElement(array, "mxPoint", {"x": str(px), "y": str(py)}) + return cell_id + + def save(self, output: Path) -> None: + ET.indent(self.mxfile, space=" ") + output.parent.mkdir(parents=True, exist_ok=True) + ET.ElementTree(self.mxfile).write(output, encoding="utf-8", xml_declaration=True) + + +def rect_style( + fill: str = WHITE, + stroke: str = RULE, + *, + font: str = INK, + size: int = 24, + bold: bool = False, + rounded: int = 1, + stroke_width: float = 1.8, + align: str = "center", + dashed: bool = False, + extra: str = "", +) -> str: + items = [ + f"rounded={rounded}", + "arcSize=10", + "whiteSpace=wrap", + "html=1", + "convertToSvg=1", + f"fillColor={fill}", + f"strokeColor={stroke}", + f"strokeWidth={stroke_width}", + f"fontColor={font}", + "fontFamily=Helvetica", + f"fontSize={size}", + f"fontStyle={1 if bold else 0}", + f"align={align}", + "verticalAlign=middle", + "spacing=4", + ] + if dashed: + items.extend(["dashed=1", "dashPattern=8 6"]) + if extra: + items.append(extra.rstrip(";")) + return ";".join(items) + ";" + + +def text_style(size: int = 24, *, font: str = INK, bold: bool = False, align: str = "left") -> str: + return ( + "text;html=1;convertToSvg=1;strokeColor=none;fillColor=none;whiteSpace=wrap;" + f"fontFamily=Helvetica;fontSize={size};fontColor={font};" + f"fontStyle={1 if bold else 0};align={align};verticalAlign=middle;spacing=0;" + ) + + +def panel_title(d: Diagram, letter: str, title: str, x: int, y: int, width: int) -> None: + d.vertex( + letter, + x, + y, + 46, + 46, + rect_style(INK, INK, font=WHITE, size=30, bold=True, extra="ellipse"), + cell_id=f"panel_{letter}", + ) + d.vertex(title, x + 62, y - 2, width, 50, text_style(32, bold=True), cell_id=f"title_{letter}") + + +def add_separator(d: Diagram, x: int, y: int, w: int, h: int = 2) -> None: + d.vertex("", x, y, w, h, rect_style(RULE, RULE, rounded=0, stroke_width=0)) + + +def token_row( + d: Diagram, + prefix: str, + x: int, + y: int, + labels: list[str], + kinds: list[str], + *, + cell_w: int = 56, + cell_h: int = 48, + gap: int = 8, + parent: str = "1", + font_size: int = 24, +) -> list[str]: + palette = { + "plain": (WHITE, RULE, INK), + "shared": (SHARED_FILL, SHARED, SHARED_DARK), + "anchor": (AMBER_FILL, AMBER, AMBER), + "mask": (GOOD_FILL, GOOD, GOOD), + "frozen": (FROZEN_FILL, FROZEN, FROZEN_DARK), + "good": (GOOD_FILL, GOOD, GOOD), + "correct": (AMBER_FILL, AMBER, AMBER), + } + ids: list[str] = [] + for idx, (label, kind) in enumerate(zip(labels, kinds)): + fill, stroke, font = palette[kind] + ids.append( + d.vertex( + label, + x + idx * (cell_w + gap), + y, + cell_w, + cell_h, + rect_style(fill, stroke, font=font, size=font_size, bold=True), + cell_id=f"{prefix}_{idx}", + parent=parent, + ) + ) + return ids + + +def document_card(d: Diagram, label: str, x: int, y: int, cell_id: str) -> str: + card = d.vertex( + label, + x, + y, + 120, + 78, + rect_style( + WHITE, + RULE, + size=19, + bold=True, + extra="container=1;pointerEvents=0;verticalAlign=top;spacingTop=22", + ), + cell_id=cell_id, + ) + d.vertex("", 0, 0, 120, 9, rect_style(SHARED, SHARED, rounded=0, stroke_width=0), parent=card) + d.vertex("", 28, 60, 64, 3, rect_style(RULE, RULE, rounded=0, stroke_width=0), parent=card) + return card + + +def frozen_target(d: Diagram, x: int, y: int, w: int, h: int, cell_id: str) -> str: + box = d.vertex( + "", + x, + y, + w, + h, + rect_style(FROZEN_FILL, FROZEN, extra="container=1;pointerEvents=0"), + cell_id=cell_id, + ) + d.vertex("Frozen target", 12, 10, w - 24, 30, text_style(22, font=FROZEN_DARK, bold=True, align="center"), parent=box) + for idx in range(3): + d.vertex( + "", + 28, + 52 + idx * 16, + w - 56, + 8, + rect_style(WHITE, RULE, rounded=0, stroke_width=1), + parent=box, + ) + return box + + +def feature_stack(d: Diagram, x: int, y: int) -> str: + stack = d.vertex( + "", + x, + y, + 190, + 92, + rect_style(WHITE, RULE, extra="container=1;pointerEvents=0"), + cell_id="frozen_feature_stack", + ) + d.vertex("Frozen features", 10, 8, 170, 28, text_style(20, font=FROZEN_DARK, bold=True, align="center"), parent=stack) + for idx, width in enumerate([150, 136, 122]): + d.vertex( + "", + 20, + 46 + idx * 13, + width, + 7, + rect_style(FROZEN_FILL, FROZEN, rounded=0, stroke_width=1), + parent=stack, + ) + return stack + + +def draw_panel_a(d: Diagram) -> dict[str, str]: + panel_title(d, "A", "Construct training signals", 40, 28, 540) + d.vertex("TRAINING GROUPS", 40, 88, 400, 34, text_style(23, font=MUTED, bold=True)) + + cards = [ + document_card(d, "Domain 1", 40, 130, "domain_1"), + document_card(d, "Domain 2", 200, 130, "domain_2"), + document_card(d, "⋯", 360, 130, "domain_mid"), + document_card(d, "Domain N", 520, 130, "domain_n"), + ] + pair = d.vertex( + "Request–response
pair (x, y)", + 90, + 250, + 280, + 78, + rect_style(WHITE, RULE, size=22, bold=True), + cell_id="xy_pair", + ) + assignment = d.vertex( + "Prompt-only
training label d", + 400, + 250, + 250, + 78, + rect_style(ROUTE_FILL, ROUTE, font=ROUTE_DARK, size=21, bold=True, stroke_width=2.0), + cell_id="offline_assignment", + ) + for idx, card in enumerate(cards): + d.edge(card, pair, width=1.5, exit_xy=(0.5, 1), entry_xy=((idx + 1) / 5, 0)) + d.edge(pair, assignment, color=ROUTE, width=2.0, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + + d.vertex("Response y", 40, 360, 130, 32, text_style(22, font=MUTED, bold=True)) + response = token_row( + d, + "response", + 210, + 350, + ["y1", "y2", "ya", "y4", "⋯"], + ["plain", "plain", "anchor", "plain", "plain"], + cell_w=60, + gap=10, + font_size=24, + ) + d.edge(pair, response[0], width=1.8, exit_xy=(0.5, 1), entry_xy=(0.5, 0), points=[(230, 340), (240, 340)]) + d.vertex("sampled anchor", 340, 405, 150, 28, text_style(20, font=AMBER, bold=True, align="center")) + + masked_box = d.vertex( + "", + 250, + 440, + 390, + 112, + rect_style(WHITE, GOOD, stroke_width=1.8, extra="container=1;pointerEvents=0"), + cell_id="masked_candidate_block", + ) + d.vertex("Masked candidate block", 15, 10, 360, 30, text_style(22, font=GOOD, bold=True, align="center"), parent=masked_box) + token_row( + d, + "masked", + 28, + 52, + ["ya", "[M]", "[M]", "[M]"], + ["anchor", "mask", "mask", "mask"], + cell_w=68, + gap=12, + parent=masked_box, + font_size=24, + ) + d.edge(response[2], masked_box, color=AMBER, width=2.0, exit_xy=(0.5, 1), entry_xy=(0.33, 0)) + + target = frozen_target(d, 40, 560, 180, 110, "frozen_target_train") + features = feature_stack(d, 250, 570) + z_port = d.vertex( + "Target context
zt", + 480, + 555, + 170, + 54, + rect_style(SHARED_FILL, SHARED, font=SHARED_DARK, size=21, bold=True), + cell_id="train_zt_port", + ) + q_port = d.vertex( + "Prompt features
q(x)", + 480, + 620, + 170, + 54, + rect_style(FROZEN_FILL, FROZEN, font=FROZEN_DARK, size=21, bold=True), + cell_id="train_q_port", + ) + d.edge(pair, target, color=FROZEN_DARK, width=1.7, exit_xy=(0, 0.55), entry_xy=(0.25, 0), points=[(20, 295), (20, 540)]) + d.edge(target, features, color=FROZEN_DARK, width=1.7, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(features, z_port, color=SHARED, width=1.8, exit_xy=(1, 0.35), entry_xy=(0, 0.5)) + d.edge(features, q_port, color=FROZEN_DARK, width=1.6, dashed=True, exit_xy=(1, 0.75), entry_xy=(0, 0.5)) + return {"assignment": assignment, "z": z_port, "q": q_port} + + +def mlp_bank(d: Diagram, x: int, y: int) -> tuple[str, str]: + bank = d.vertex( + "", + x, + y, + 300, + 260, + rect_style(WHITE, RULE, extra="container=1;pointerEvents=0"), + cell_id="train_mlp_bank", + ) + d.vertex("MLP bank", 15, 10, 270, 34, text_style(25, bold=True, align="center"), parent=bank) + labels = ["MLPℓ,1", "MLPℓ,2", "MLPℓ,d", "⋯", "MLPℓ,N"] + selected = "" + for idx, label in enumerate(labels): + active = idx == 2 + row = d.vertex( + label, + 18, + 54 + idx * 38, + 264, + 30, + rect_style( + MLP_FILL if active else MLP_FAINT, + MLP if active else RULE, + font=MLP_DARK if active else MUTED, + size=22, + bold=active, + stroke_width=2.2 if active else 1.1, + ), + cell_id=f"train_mlp_{idx + 1}", + parent=bank, + ) + if active: + selected = row + return bank, selected + + +def draw_panel_b(d: Diagram) -> dict[str, str]: + panel_title(d, "B", "Jointly train MoS", 750, 28, 520) + d.vertex("ONE EXPANDED DRAFT LAYER", 750, 88, 500, 34, text_style(23, font=SHARED_DARK, bold=True)) + + h_in = d.vertex("hℓ", 760, 250, 90, 66, rect_style(WHITE, RULE, size=30, bold=True), cell_id="train_h_in") + norm_1 = d.vertex("Norm", 890, 250, 110, 66, rect_style(SHARED_FILL, SHARED, font=SHARED_DARK, size=25, bold=True), cell_id="train_norm_1") + attn = d.vertex( + "Shared
attention", + 1040, + 215, + 210, + 136, + rect_style(SHARED_FILL, SHARED, font=SHARED_DARK, size=27, bold=True, stroke_width=2.2), + cell_id="train_attention", + ) + plus_1 = d.vertex("+", 1280, 257, 52, 52, rect_style(WHITE, INK, size=32, bold=True, extra="ellipse"), cell_id="train_plus_1") + norm_2 = d.vertex("Norm", 1370, 250, 100, 66, rect_style(SHARED_FILL, SHARED, font=SHARED_DARK, size=25, bold=True), cell_id="train_norm_2") + bank, selected = mlp_bank(d, 1500, 135) + plus_2 = d.vertex("+", 1840, 257, 52, 52, rect_style(WHITE, INK, size=32, bold=True, extra="ellipse"), cell_id="train_plus_2") + h_out = d.vertex("hℓ+1", 1930, 250, 100, 66, rect_style(WHITE, RULE, size=28, bold=True), cell_id="train_h_out") + + z_tokens = token_row( + d, + "zt_train", + 790, + 145, + ["zt1", "zt2", "⋯"], + ["shared", "shared", "shared"], + cell_w=56, + gap=8, + font_size=22, + ) + projection = d.vertex( + "Target-context
projection", + 1040, + 130, + 210, + 66, + rect_style(SHARED_FILL, SHARED, font=SHARED_DARK, size=23, bold=True), + cell_id="train_projection", + ) + selector = d.vertex( + "Training label d", + 1280, + 120, + 180, + 44, + rect_style(ROUTE_FILL, ROUTE, font=ROUTE_DARK, size=21, bold=True, stroke_width=2.0), + cell_id="training_selector", + ) + + d.edge(h_in, norm_1, width=2.2, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(norm_1, attn, width=2.2, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(attn, plus_1, width=2.2, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(plus_1, norm_2, width=2.2, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(norm_2, selected, color=MLP, width=2.6, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(selected, plus_2, color=MLP, width=2.6, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(plus_2, h_out, width=2.2, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(z_tokens[-1], projection, color=SHARED, width=1.9, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(projection, attn, color=SHARED, width=1.9, exit_xy=(0.5, 1), entry_xy=(0.5, 0)) + d.edge( + selector, + selected, + color=ROUTE, + width=1.9, + dashed=True, + exit_xy=(1, 0.5), + entry_xy=(0, 0.5), + points=[(1480, 142), (1480, 280)], + ) + + d.edge(h_in, plus_1, width=1.7, exit_xy=(0.5, 1), entry_xy=(0.5, 1), points=[(805, 380), (1305, 380)]) + d.vertex("residual", 940, 390, 120, 28, text_style(20, font=MUTED, align="center")) + d.edge(plus_1, plus_2, width=1.7, exit_xy=(0.5, 1), entry_xy=(0.5, 1), points=[(1305, 420), (1865, 420)]) + d.vertex("residual", 1540, 388, 120, 28, text_style(20, font=MUTED, align="center")) + + d.vertex("Same d across layers", 760, 445, 240, 34, text_style(22, font=MLP_DARK, bold=True)) + group_boxes = [] + for idx, (x, label) in enumerate([(1060, "MLP1,d"), (1270, "MLP2,d"), (1550, "MLPL,d")]): + group_boxes.append( + d.vertex( + label, + x, + 438, + 160, + 52, + rect_style(MLP_FILL, MLP, font=MLP_DARK, size=22, bold=True, stroke_width=2.0), + cell_id=f"same_group_{idx}", + ) + ) + dots = d.vertex("⋯", 1455, 445, 60, 34, text_style(28, font=MLP_DARK, bold=True, align="center")) + d.edge(group_boxes[0], group_boxes[1], color=MLP, width=2.2, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(group_boxes[1], dots, color=MLP, width=2.2, exit_xy=(1, 0.5), entry_xy=(0, 0.5), end_arrow="none") + d.edge(dots, group_boxes[2], color=MLP, width=2.2, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + + init = d.vertex( + "", + 2050, + 80, + 310, + 160, + rect_style(SURFACE, RULE, extra="container=1;pointerEvents=0"), + cell_id="initialization_inset", + ) + d.vertex("Initialization", 15, 10, 280, 30, text_style(24, bold=True, align="center"), parent=init) + d.vertex("Public DFlash or trained generalist", 15, 50, 280, 30, text_style(19, font=MUTED, align="center"), parent=init) + d.vertex("↓", 135, 80, 40, 20, text_style(22, font=MUTED, bold=True, align="center"), parent=init) + d.vertex("Shared modules +
copied MLP groups", 15, 108, 280, 42, text_style(20, font=MLP_DARK, bold=True, align="center"), parent=init) + + logits = d.vertex("Parallel predictions", 2070, 280, 270, 52, rect_style(MLP_FAINT, MLP, font=MLP_DARK, size=21, bold=True), cell_id="block_logits") + block_loss = d.vertex("Block-prediction loss", 2070, 360, 270, 52, rect_style(WHITE, MLP, font=MLP_DARK, size=21, bold=True, stroke_width=2.0), cell_id="block_loss") + block_update = d.vertex("Updates shared modules
and selected MLP", 2070, 440, 270, 62, rect_style(MLP_FILL, MLP, font=MLP_DARK, size=19, bold=True, stroke_width=1.8), cell_id="block_update") + d.edge(h_out, logits, color=MLP, width=2.0, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(logits, block_loss, color=MLP, width=2.0, exit_xy=(0.5, 1), entry_xy=(0.5, 0)) + d.edge(block_loss, block_update, color=MLP, width=2.0, exit_xy=(0.5, 1), entry_xy=(0.5, 0)) + + d.vertex("ROUTER SUPERVISION", 760, 535, 300, 30, text_style(22, font=ROUTE_DARK, bold=True)) + detached = d.vertex("Prompt features
(no gradient)", 760, 575, 230, 64, rect_style(FROZEN_FILL, FROZEN, font=FROZEN_DARK, size=19, bold=True, dashed=True), cell_id="router_detached_features") + pool = d.vertex("Mean pool", 1020, 580, 130, 58, rect_style(ROUTE_FILL, ROUTE, font=ROUTE_DARK, size=21, bold=True), cell_id="router_pool") + router = d.vertex("Request router", 1190, 580, 170, 58, rect_style(ROUTE_FILL, ROUTE, font=ROUTE_DARK, size=20, bold=True), cell_id="request_router_train") + route_loss = d.vertex("Routing loss", 1390, 580, 150, 58, rect_style(WHITE, ROUTE, font=ROUTE_DARK, size=20, bold=True, stroke_width=2.0), cell_id="route_loss") + router_update = d.vertex("Router update", 1570, 580, 180, 58, rect_style(ROUTE_FILL, ROUTE, font=ROUTE_DARK, size=20, bold=True), cell_id="router_update") + d.edge(detached, pool, color=ROUTE, width=1.9, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(pool, router, color=ROUTE, width=1.9, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(router, route_loss, color=ROUTE, width=1.9, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(route_loss, router_update, color=ROUTE, width=1.9, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + + checkpoint = d.vertex( + "", + 2070, + 535, + 270, + 125, + rect_style(WHITE, INK, stroke_width=2.0, extra="container=1;pointerEvents=0"), + cell_id="mos_checkpoint", + ) + d.vertex("MoS checkpoint", 15, 12, 240, 32, text_style(23, bold=True, align="center"), parent=checkpoint) + d.vertex("Shared modules · MLP bank
Request router", 15, 52, 240, 54, text_style(19, font=MUTED, align="center"), parent=checkpoint) + d.edge(block_update, checkpoint, color=MLP, width=1.8, exit_xy=(0.5, 1), entry_xy=(0.65, 0)) + d.edge(router_update, checkpoint, color=ROUTE, width=1.8, exit_xy=(1, 0.5), entry_xy=(0, 0.75), points=[(1870, 609), (1870, 630)]) + return {"checkpoint": checkpoint} + + +def inference_layer(d: Diagram, prefix: str, x: int, label: str, selected_label: str, *, parent: str) -> tuple[str, str, str]: + layer = d.vertex( + "", + x, + 62, + 340, + 170, + rect_style(WHITE, RULE, extra="container=1;pointerEvents=0"), + cell_id=f"{prefix}_layer", + parent=parent, + ) + d.vertex(label, 15, 10, 310, 34, text_style(25, bold=True, align="center"), parent=layer) + shared = d.vertex( + "Shared
modules", + 18, + 60, + 165, + 88, + rect_style(SHARED_FILL, SHARED, font=SHARED_DARK, size=22, bold=True, stroke_width=2.0), + cell_id=f"{prefix}_shared", + parent=layer, + ) + selected = d.vertex( + selected_label, + 205, + 64, + 115, + 56, + rect_style(MLP_FILL, MLP, font=MLP_DARK, size=20, bold=True, stroke_width=2.2), + cell_id=f"{prefix}_selected", + parent=layer, + ) + d.vertex("", 215, 132, 95, 7, rect_style(MLP_FAINT, RULE, rounded=0, stroke_width=1.0), cell_id=f"{prefix}_inactive_1", parent=layer) + d.vertex("", 215, 145, 95, 7, rect_style(MLP_FAINT, RULE, rounded=0, stroke_width=1.0), cell_id=f"{prefix}_inactive_2", parent=layer) + d.edge(shared, selected, color=MLP, width=2.3, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + return layer, shared, selected + + +def candidate_block(d: Diagram, x: int, y: int) -> tuple[str, str]: + box = d.vertex( + "", + x, + y, + 340, + 210, + rect_style(WHITE, MLP, stroke_width=2.0, extra="container=1;pointerEvents=0"), + cell_id="candidate_block", + ) + d.vertex("Candidate block (parallel)", 15, 12, 310, 34, text_style(21, font=MLP_DARK, bold=True, align="center"), parent=box) + tokens = token_row( + d, + "candidate_tokens", + 18, + 70, + ["t1", "t2", "t3", "⋯", "t16"], + ["plain"] * 5, + cell_w=48, + gap=10, + parent=box, + font_size=21, + ) + bus = d.vertex("", 35, 166, 270, 3, rect_style(MLP, MLP, rounded=0, stroke_width=0), cell_id="candidate_parallel_bus", parent=box) + for idx in [0, 2, 4]: + d.edge(bus, tokens[idx], color=MLP, width=1.6, exit_xy=((idx + 1) / 6, 0), entry_xy=(0.5, 1)) + return box, bus + + +def verification_box(d: Diagram, x: int, y: int) -> str: + box = d.vertex( + "", + x, + y, + 320, + 230, + rect_style(FROZEN_FILL, FROZEN, stroke_width=2.0, extra="container=1;pointerEvents=0"), + cell_id="target_verification", + ) + d.vertex("Target verification", 15, 12, 290, 36, text_style(24, font=FROZEN_DARK, bold=True, align="center"), parent=box) + token_row( + d, + "verify_tokens", + 32, + 66, + ["t1", "t2", "t3", "t4"], + ["good", "good", "good", "correct"], + cell_w=54, + gap=10, + parent=box, + font_size=21, + ) + d.vertex("✓ ✓ ✓", 40, 122, 190, 30, text_style(24, font=GOOD, bold=True, align="center"), parent=box) + d.vertex("fix", 245, 122, 45, 30, text_style(19, font=AMBER, bold=True, align="center"), parent=box) + d.vertex("Accept matching prefix", 25, 164, 270, 28, text_style(20, font=GOOD, bold=True, align="center"), parent=box) + d.vertex("Correct first mismatch", 25, 198, 270, 28, text_style(20, font=AMBER, bold=True, align="center"), parent=box) + return box + + +def draw_panel_c(d: Diagram, checkpoint: str) -> None: + panel_title(d, "C", "Request-routed speculative decoding", 40, 730, 760) + d.vertex("ROUTE ONCE", 40, 792, 220, 32, text_style(23, font=ROUTE_DARK, bold=True)) + + prompt = d.vertex("Prompt x", 40, 845, 160, 68, rect_style(WHITE, RULE, size=26, bold=True), cell_id="infer_prompt") + prefill = d.vertex("Frozen target
prompt pass", 240, 835, 230, 88, rect_style(FROZEN_FILL, FROZEN, font=FROZEN_DARK, size=22, bold=True), cell_id="target_prefill") + feature_box = d.vertex( + "Prompt features", + 510, + 835, + 180, + 88, + rect_style(FROZEN_FILL, FROZEN, font=FROZEN_DARK, size=20, bold=True, extra="verticalAlign=top;spacingTop=10"), + cell_id="prompt_features", + ) + for idx, width in enumerate([130, 110, 90]): + d.vertex( + "", + 25, + 48 + idx * 10, + width, + 5, + rect_style(WHITE, RULE, rounded=0, stroke_width=1), + cell_id=f"prompt_feature_strip_{idx}", + parent=feature_box, + ) + pool = d.vertex("Mean pool", 730, 845, 150, 68, rect_style(ROUTE_FILL, ROUTE, font=ROUTE_DARK, size=22, bold=True), cell_id="infer_mean_pool") + router = d.vertex("Request router", 920, 845, 190, 68, rect_style(ROUTE_FILL, ROUTE, font=ROUTE_DARK, size=22, bold=True), cell_id="infer_request_router") + choice = d.vertex( + "Predicted
group dpred", + 1150, + 835, + 130, + 88, + rect_style(ROUTE_FILL, ROUTE, font=ROUTE_DARK, size=20, bold=True, stroke_width=2.2), + cell_id="route_choice", + ) + lock = d.vertex( + "LOCK", + 1320, + 845, + 90, + 68, + rect_style(WHITE, ROUTE, font=ROUTE_DARK, size=20, bold=True, stroke_width=2.2), + cell_id="route_lock", + ) + d.edge(prompt, prefill, width=2.1, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(prefill, feature_box, color=FROZEN_DARK, width=1.8, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(feature_box, pool, color=ROUTE, width=1.9, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(pool, router, color=ROUTE, width=1.9, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(router, choice, color=ROUTE, width=1.9, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(choice, lock, color=ROUTE, width=1.9, exit_xy=(1, 0.5), entry_xy=(0, 0.5), end_arrow="none") + + d.vertex("One decision per request", 1460, 825, 360, 34, text_style(25, font=ROUTE_DARK, bold=True)) + d.vertex("cycle 1 — cycle 2 — ⋯ — cycle T", 1460, 866, 520, 32, text_style(23, font=ROUTE_DARK)) + d.vertex("Fixed for all layers and cycles", 1460, 904, 420, 30, text_style(22, font=ROUTE_DARK, bold=True)) + + loaded = d.vertex("Trained MoS checkpoint", 2110, 835, 250, 72, rect_style(SURFACE, RULE, font=MUTED, size=21, bold=True), cell_id="loaded_checkpoint") + d.edge(checkpoint, loaded, color=MUTED, width=1.6, dashed=True, exit_xy=(0.75, 1), entry_xy=(0.75, 0), points=[(2300, 690), (2300, 810)]) + + d.vertex("DATA PATH", 40, 950, 180, 32, text_style(23, font=MUTED, bold=True)) + context = d.vertex("Target context
zt", 40, 1000, 190, 70, rect_style(SHARED_FILL, SHARED, font=SHARED_DARK, size=22, bold=True), cell_id="infer_context") + masked = d.vertex("Masked block
[M] [M] [M]", 40, 1110, 190, 70, rect_style(GOOD_FILL, GOOD, font=GOOD, size=21, bold=True), cell_id="infer_masked") + d.edge( + prefill, + context, + color=SHARED, + width=1.8, + exit_xy=(0.05, 1), + entry_xy=(1, 0.5), + points=[(250, 975), (250, 1035)], + ) + + path = d.vertex( + "", + 290, + 950, + 1220, + 280, + rect_style(SURFACE, INK, stroke_width=2.0, extra="container=1;pointerEvents=0"), + cell_id="selected_path", + ) + d.vertex("Selected MoS path", 20, 12, 300, 34, text_style(27, bold=True), parent=path) + d.vertex("fixed group dpred for the whole request", 670, 14, 520, 30, text_style(21, font=MLP_DARK, bold=True, align="right"), parent=path) + layer_1, shared_1, selected_1 = inference_layer(d, "infer_1", 20, "Layer 1", "MLP1", parent=path) + layer_2, shared_2, selected_2 = inference_layer(d, "infer_2", 420, "Layer 2", "MLP2", parent=path) + layer_l, shared_l, selected_l = inference_layer(d, "infer_l", 820, "Layer L", "MLPL", parent=path) + d.edge(context, shared_1, color=SHARED, width=2.0, exit_xy=(1, 0.5), entry_xy=(0, 0.3), points=[(260, 1035), (290, 1035)]) + d.edge(masked, shared_1, color=GOOD, width=2.0, exit_xy=(1, 0.5), entry_xy=(0, 0.75), points=[(260, 1145), (290, 1145)]) + d.edge(selected_1, shared_2, color=MLP, width=2.6, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.edge(selected_2, shared_l, color=MLP, width=2.6, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + d.vertex("Other MLP groups are not evaluated or merged", 250, 242, 720, 26, text_style(20, font=MUTED, bold=True, align="center"), parent=path) + + candidate, candidate_bus = candidate_block(d, 1550, 970) + d.edge(selected_l, candidate_bus, color=MLP, width=2.6, exit_xy=(1, 0.5), entry_xy=(0, 0.5), points=[(1530, 1110)]) + verify = verification_box(d, 1910, 950) + d.edge(candidate, verify, width=2.1, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + response = d.vertex("Response", 2250, 1025, 140, 72, rect_style(WHITE, RULE, size=21, bold=True), cell_id="response_output") + d.edge(verify, response, width=2.1, exit_xy=(1, 0.5), entry_xy=(0, 0.5)) + + d.vertex("Accepted or corrected tokens start the next verification cycle", 560, 1260, 900, 32, text_style(22, font=MUTED, bold=True, align="center"), cell_id="cycle_note") + d.edge(verify, context, color=FROZEN_DARK, width=1.6, exit_xy=(0.5, 1), entry_xy=(0, 0.5), points=[(2070, 1310), (20, 1310), (20, 1035)]) + + roles = [ + (SHARED_FILL, SHARED, "shared"), + (MLP_FILL, MLP, "selected MLP"), + (ROUTE_FILL, ROUTE, "routing"), + (FROZEN_FILL, FROZEN, "frozen / inactive"), + ] + for idx, (fill, stroke, label) in enumerate(roles): + x = 40 + idx * 260 + d.vertex("", x, 1360, 34, 24, rect_style(fill, stroke, rounded=0, stroke_width=1.6), cell_id=f"legend_swatch_{idx}") + d.vertex(label, x + 46, 1354, 200, 36, text_style(21, font=MUTED, bold=True), cell_id=f"legend_label_{idx}") + d.vertex("Exact verification preserves the target distribution", 1500, 1354, 860, 36, text_style(22, font=MUTED, bold=True, align="right"), cell_id="exactness_note") + + +def build() -> Diagram: + diagram = Diagram() + add_separator(diagram, 700, 20, 2, 650) + add_separator(diagram, 30, 700, 2240, 2) + add_separator(diagram, 2330, 700, 40, 2) + draw_panel_a(diagram) + panel_b = draw_panel_b(diagram) + draw_panel_c(diagram, panel_b["checkpoint"]) + return diagram + + +def main() -> None: + args = parse_args() + build().save(args.output) + + +if __name__ == "__main__": + main() diff --git a/recipes/plotting/v1/fig_5arm_per_domain.py b/recipes/plotting/v1/fig_5arm_per_domain.py new file mode 100644 index 0000000000000000000000000000000000000000..52d6f92f3430854375f25da380866b647f651d58 --- /dev/null +++ b/recipes/plotting/v1/fig_5arm_per_domain.py @@ -0,0 +1,49 @@ +import matplotlib +matplotlib.use("Agg") +import matplotlib.pyplot as plt +import numpy as np + +domains = ["code", "math", "factual_qa", "creative_writing", "general"] +nan = np.nan +series = [ + ("D0 (base)", [2.44, 3.38, 2.23, 2.11, 2.43], "#cfd8dc"), + ("small-data spec.", [3.16, 4.49, 2.68, 2.54, 2.91], "#90a4ae"), + ("warm spec.", [3.20, 4.68, 2.86, 2.67, 3.06], "#607d8b"), + ("Generalist", [3.205, 4.698, 2.814, 2.651, 3.086], "#1565c0"), + ("big-data spec. 250k",[3.340, 5.190, 2.932, 2.861, 3.117], "#2e7d32"), +] +x = np.arange(len(domains)); n = len(series); w = 0.16 +fig, ax = plt.subplots(figsize=(10.5, 5.2)) +for i, (name, vals, c) in enumerate(series): + off = (i - (n-1)/2) * w + bars = ax.bar(x + off, vals, w, label=name, color=c) + for b, v in zip(bars, vals): + if not np.isnan(v): + ax.text(b.get_x()+b.get_width()/2, v+0.02, f"{v:.2f}", + ha="center", va="bottom", fontsize=7, rotation=90) +# mark big-data wins (code, math) +ax.annotate("+0.135", xy=(x[0]+2*w, 3.34), xytext=(x[0]+2*w, 3.72), + ha="center", fontsize=10, fontweight="bold", + arrowprops=dict(arrowstyle="->", color="#2e7d32", lw=1.4)) +ax.annotate("+0.49", xy=(x[1]+2*w, 5.19), xytext=(x[1]+2*w, 5.60), + ha="center", fontsize=10, fontweight="bold", + arrowprops=dict(arrowstyle="->", color="#2e7d32", lw=1.4)) +ax.annotate("+0.118", xy=(x[2]+2*w, 2.93), xytext=(x[2]+2*w, 3.30), + ha="center", fontsize=10, fontweight="bold", + arrowprops=dict(arrowstyle="->", color="#2e7d32", lw=1.4)) +ax.annotate("+0.21", xy=(x[3]+2*w, 2.86), xytext=(x[3]+2*w, 3.25), + ha="center", fontsize=10, fontweight="bold", + arrowprops=dict(arrowstyle="->", color="#2e7d32", lw=1.4)) +ax.annotate("+0.03", xy=(x[4]+2*w, 3.12), xytext=(x[4]+2*w, 3.50), + ha="center", fontsize=9, color="#555", + arrowprops=dict(arrowstyle="->", color="#999", lw=1.0)) +ax.set_xticks(x); ax.set_xticklabels(domains) +ax.set_ylabel("held-out accept length (AL)") +ax.set_ylim(2.0, 5.85) +ax.set_title("Per-domain AL: D0 / small-data / warm / Generalist / big-data(250k) specialist\n" + "(big-data 5/5: math +0.49, creative_writing +0.21, code +0.135, factual_qa +0.118, general +0.03)") +ax.legend(ncol=5, loc="upper center", fontsize=8.5, framealpha=0.9) +ax.grid(axis="y", ls=":", alpha=0.5) +fig.tight_layout() +fig.savefig("reasonmix_5arm_per_domain.png", dpi=130) +print("OK") diff --git a/recipes/plotting/v1/fig_code250k_vs_gen.py b/recipes/plotting/v1/fig_code250k_vs_gen.py new file mode 100644 index 0000000000000000000000000000000000000000..25a5c603e85c1a3c2d6d853d7026800654eada11 --- /dev/null +++ b/recipes/plotting/v1/fig_code250k_vs_gen.py @@ -0,0 +1,33 @@ +import matplotlib +matplotlib.use("Agg") +import matplotlib.pyplot as plt +import numpy as np + +epochs = ["ep0", "ep1", "ep2", "ep3"] +gen = [3.072, 3.150, 3.205, 3.204] # generalist on code held-out +spec = [3.191, 3.268, 3.340, 3.339] # code250k specialist on code held-out +x = np.arange(len(epochs)); w = 0.38 + +fig, ax = plt.subplots(figsize=(7.2, 4.6)) +b1 = ax.bar(x - w/2, gen, w, label="generalist (mixed 250k, 91k code)", color="#9aa7b8") +b2 = ax.bar(x + w/2, spec, w, label="code specialist (from D0, 250k code)", color="#2e7d32") + +ax.axhline(3.205, ls="--", lw=1.2, color="#9aa7b8") +ax.axhline(3.340, ls="--", lw=1.2, color="#2e7d32") +ax.annotate("", xy=(3.34, 3.340), xytext=(3.34, 3.205), + arrowprops=dict(arrowstyle="<->", color="black", lw=1.3)) +ax.text(3.0, (3.205+3.340)/2, "+0.135", fontsize=12, fontweight="bold", va="center") + +for b in list(b1)+list(b2): + ax.text(b.get_x()+b.get_width()/2, b.get_height()+0.006, + f"{b.get_height():.3f}", ha="center", va="bottom", fontsize=8.5) + +ax.set_xticks(x); ax.set_xticklabels(epochs) +ax.set_ylim(3.0, 3.42) +ax.set_ylabel("code held-out accept length (AL)") +ax.set_title("code: 250k single-domain specialist vs generalist (per-epoch, held-out=89)") +ax.legend(loc="lower right", fontsize=9) +ax.grid(axis="y", ls=":", alpha=0.5) +fig.tight_layout() +fig.savefig("code250k_vs_gen.png", dpi=130) +print("OK") diff --git a/recipes/plotting/v1/fig_exp1_forgetting.py b/recipes/plotting/v1/fig_exp1_forgetting.py new file mode 100644 index 0000000000000000000000000000000000000000..ca3ef066a2dd39e9d55e364930c879a0e2a4cb11 --- /dev/null +++ b/recipes/plotting/v1/fig_exp1_forgetting.py @@ -0,0 +1,58 @@ +#!/usr/bin/env python3 +# Exp 1 forgetting trajectories: continue-train the GENERALIST on ONE domain (math / code), +# full-param (B) vs MLP-only/frozen-attn (A), baseline (k5clean) data, eval every epoch on all 5 domains. +# Shows: target domain rises; OTHER domains drop (forgetting) EVEN WITH MLP-only -> forgetting lives in the MLP. +import matplotlib; matplotlib.use("Agg") +import matplotlib.pyplot as plt +import numpy as np + +EP = list(range(6)) +DOMS = ["code", "math", "factual_qa", "creative_writing", "general"] +GEN = {"code": 3.209, "math": 4.698, "factual_qa": 2.814, "creative_writing": 2.636, "general": 3.049} + +# [trained_domain][arm][eval_domain] = per-epoch AL (ep0..ep5) +DATA = { + "math": { + "B": {"code":[3.183,3.163,3.143,3.166,3.161,3.157],"math":[4.777,4.758,4.843,4.915,4.907,4.912], + "factual_qa":[2.844,2.741,2.760,2.743,2.758,2.809],"creative_writing":[2.606,2.618,2.567,2.599,2.584,2.579], + "general":[3.040,3.029,3.031,3.033,3.064,3.048]}, + "A": {"code":[3.177,3.183,3.153,3.178,3.169,3.188],"math":[4.748,4.781,4.798,4.800,4.842,4.847], + "factual_qa":[2.834,2.765,2.766,2.747,2.830,2.769],"creative_writing":[2.663,2.596,2.599,2.628,2.629,2.606], + "general":[3.067,3.046,3.039,3.075,3.090,3.056]}, + }, + "code": { + "B": {"code":[3.208,3.239,3.296,3.303,3.329,3.338],"math":[4.630,4.646,4.583,4.574,4.559,4.578], + "factual_qa":[2.781,2.739,2.820,2.747,2.753,2.744],"creative_writing":[2.602,2.569,2.572,2.590,2.633,2.598], + "general":[3.017,3.029,3.030,3.021,3.022,3.048]}, + "A": {"code":[3.226,3.261,3.263,3.308,3.304,3.334],"math":[4.665,4.613,4.580,4.600,4.575,4.585], + "factual_qa":[2.815,2.839,2.766,2.767,2.746,2.812],"creative_writing":[2.630,2.631,2.622,2.617,2.613,2.609], + "general":[3.055,3.042,3.048,3.065,3.042,3.033]}, + }, +} + +fig, axes = plt.subplots(2, 5, figsize=(18, 6.8), sharex=True) +for r, trained in enumerate(["math", "code"]): + for c, ed in enumerate(DOMS): + ax = axes[r][c] + is_target = (ed == trained) + ax.plot(EP, DATA[trained]["B"][ed], "o-", color="#2e7d32", lw=1.8, ms=4, label="full-param (B)") + ax.plot(EP, DATA[trained]["A"][ed], "s-", color="#e67e22", lw=1.8, ms=4, label="MLP-only / frozen-attn (A)") + ax.axhline(GEN[ed], ls="--", color="#888", lw=1.2, label="Generalist (start)") + ax.set_title(f"eval={ed}" + (" ◀ TARGET" if is_target else ""), + fontsize=9.5, fontweight=("bold" if is_target else "normal"), + color=("#1a1a1a" if is_target else "#555")) + ax.grid(alpha=0.3) + if is_target: + ax.set_facecolor("#eef7ee") + if c == 0: + ax.set_ylabel(f"train on {trained.upper()}\nheld-out AL", fontsize=10) + if r == 1: + ax.set_xlabel("epoch") + ax.tick_params(labelsize=8) +axes[0][0].legend(fontsize=7.5, loc="best") +fig.suptitle("Exp 1 — Forgetting trajectory: continue-train Generalist on ONE domain (baseline data), eval all 5 every epoch.\n" + "TARGET (green bg) rises; OTHER domains fall below Generalist (forgetting). Code→math forgetting is the same for full-param and MLP-only " + "→ forgetting lives in the MLP (router to per-domain MLP isolates it).", fontsize=11, y=1.02) +fig.tight_layout() +fig.savefig("fig_exp1_forgetting.png", dpi=135, bbox_inches="tight") +print("OK wrote /tmp/fig_exp1_forgetting.png") diff --git a/recipes/plotting/v1/fig_exp1_three_recipes.py b/recipes/plotting/v1/fig_exp1_three_recipes.py new file mode 100644 index 0000000000000000000000000000000000000000..49b02ddd13227c627284d4105c9f8c665156cc3b --- /dev/null +++ b/recipes/plotting/v1/fig_exp1_three_recipes.py @@ -0,0 +1,33 @@ +#!/usr/bin/env python3 +# Exp1: three specialist recipes (big-data / small-data / warm-MLP) vs generalist, on each domain's own data. +# One grouped bar chart. Only big-data specialist beats generalist. +import matplotlib; matplotlib.use("Agg") +import matplotlib.pyplot as plt +import numpy as np + +DOMS = ["math", "code", "fqa", "cw", "general"] +GEN = [4.698, 3.209, 2.814, 2.647, 3.086] # generalist (best single ckpt) +BIGDATA = [5.190, 3.340, 2.932, 2.861, 3.117] # big-data specialist (单域 250k), peak +SMALLDATA= [4.493, 3.162, 2.683, 2.536, 2.908] # small-data specialist (单域自然量), peak +WARM = [4.748, 3.215, 2.861, 2.671, 3.106] # warm MLP-only specialist, peak + +x = np.arange(len(DOMS)); w = 0.2 +fig, ax = plt.subplots(figsize=(11, 5.4)) +bars = [ + ax.bar(x - 1.5*w, GEN, w, label="generalist (monolithic baseline)", color="#9aa7b8"), + ax.bar(x - 0.5*w, BIGDATA, w, label="big-data specialist (250k/domain)", color="#1b5e20"), + ax.bar(x + 0.5*w, SMALLDATA, w, label="small-data specialist (natural share)", color="#c0392b"), + ax.bar(x + 1.5*w, WARM, w, label="warm specialist (MLP-only)", color="#e0a030"), +] +for bs in bars: + for b in bs: + ax.text(b.get_x()+b.get_width()/2, b.get_height()+0.02, f"{b.get_height():.2f}", + ha="center", va="bottom", fontsize=7) +ax.set_xticks(x); ax.set_xticklabels(DOMS) +ax.set_ylabel("held-out accept length (AL)") +ax.set_ylim(2.0, 5.7) +ax.set_title("Only the big-data specialist (250k/domain) beats the generalist on all 5 domains;\nsmall-data and warm specialists do not", fontsize=11) +ax.legend(loc="upper right", fontsize=9, ncol=2); ax.grid(axis="y", ls=":", alpha=0.4) +fig.tight_layout() +fig.savefig("fig_exp1_three_recipes.png", dpi=140, bbox_inches="tight") +print("OK wrote /tmp/fig_exp1_three_recipes.png") diff --git a/recipes/plotting/v1/fig_exp4_inference.py b/recipes/plotting/v1/fig_exp4_inference.py new file mode 100644 index 0000000000000000000000000000000000000000..3ceaae1607edfddd655df8a90301f75346c85e20 --- /dev/null +++ b/recipes/plotting/v1/fig_exp4_inference.py @@ -0,0 +1,51 @@ +#!/usr/bin/env python3 +# Exp 4 inference cost: (A) end-to-end single-stream tok/s merged~specialist>gen; (B) serve-step component +# breakdown (decode regime) — self_attn dominates, the lm_head "giant" is only ~6.5%. +import matplotlib; matplotlib.use("Agg") +import matplotlib.pyplot as plt +import numpy as np + +ARMS = ["generalist", "merged", "specialist"] +TOKS = [505.1, 567.3, 559.8] # single-stream tok/s (batch=1, fa3, math held-out) +AL = [4.69, 5.32, 5.24] +COL = {"generalist": "#9aa7b8", "merged": "#2e7d32", "specialist": "#b8860b"} + +fig, (axA, axB) = plt.subplots(1, 2, figsize=(12.5, 5.0)) + +# --- Panel A: end-to-end tok/s --- +x = np.arange(len(ARMS)) +bars = axA.bar(x, TOKS, width=0.6, color=[COL[a] for a in ARMS]) +for i, b in enumerate(bars): + axA.text(b.get_x()+b.get_width()/2, b.get_height()+3, f"{TOKS[i]:.0f}\nAL {AL[i]:.2f}", + ha="center", va="bottom", fontsize=9) +axA.set_xticks(x); axA.set_xticklabels(ARMS) +axA.set_ylabel("single-stream throughput (tok/s)") +axA.set_ylim(0, 640) +axA.set_title("A. End-to-end speed (math, batch=1)\nmerged 567 ≈ specialist 560, +12% over generalist 505", fontsize=10.5) +axA.text(0.5, 0.04, "merged delivers specialist-level speed at single-model per-token cost", + transform=axA.transAxes, ha="center", fontsize=8.5, style="italic", color="#444") +axA.grid(axis="y", ls=":", alpha=0.4) + +# --- Panel B: serve-step component breakdown (ms), all arms ~identical -> show one --- +comp_names = ["self_attn\n(5L)", "draft_other\n(embed/norm/\nresidual/cache)", "mlp\n(5L)", "lm_head\n(GIANT)"] +comp_ms = [2.806, 1.487, 0.641, 0.342] +comp_col = ["#c0392b", "#7f8c8d", "#2980b9", "#f1c40f"] +full = sum(comp_ms) +xb = np.arange(len(comp_names)) +bb = axB.bar(xb, comp_ms, width=0.6, color=comp_col) +for i, b in enumerate(bb): + axB.text(b.get_x()+b.get_width()/2, b.get_height()+0.03, f"{comp_ms[i]:.2f}ms\n{100*comp_ms[i]/full:.0f}%", + ha="center", va="bottom", fontsize=8.5) +axB.set_xticks(xb); axB.set_xticklabels(comp_names, fontsize=8) +axB.set_ylabel("serve-step time (ms, decode block of 16)") +axB.set_ylim(0, 3.4) +axB.set_title("B. Where the drafter step spends time (decode regime)\nself_attn dominates (53%); the lm_head 'giant' is only 6.5%", fontsize=10.5) +axB.text(0.5, 0.92, "per-forward step ≈ identical across all 3 arms (within 1.5%)", + transform=axB.transAxes, ha="center", fontsize=8.5, style="italic", color="#444") +axB.grid(axis="y", ls=":", alpha=0.4) + +fig.suptitle("Exp 4 — Merged drafter: single-model per-token cost, specialist-level throughput, negligible router (0.0002 ms/tok)", + fontsize=11.5, y=1.02) +fig.tight_layout() +fig.savefig("fig_exp4_inference.png", dpi=140, bbox_inches="tight") +print("OK wrote /tmp/fig_exp4_inference.png") diff --git a/recipes/plotting/v1/fig_exp5_method_vs_data.py b/recipes/plotting/v1/fig_exp5_method_vs_data.py new file mode 100644 index 0000000000000000000000000000000000000000..6c14dd8bfb7dc4d1fd12780c52b77d729939ddce --- /dev/null +++ b/recipes/plotting/v1/fig_exp5_method_vs_data.py @@ -0,0 +1,38 @@ +#!/usr/bin/env python3 +# Exp6 method>data: at the SAME data budget, three tiers per domain — +# naive (from D0, before our fix; LOSES gen) < generalist < CorDA-MoS warm-from-gen (OURS; beats gen). +# Shows how much our method (warm-from-gen + CorDA fusion) advances over the naive same-data attempt. +import matplotlib; matplotlib.use("Agg") +import matplotlib.pyplot as plt +import numpy as np + +# ascending by our method -> tallest (math) at the right. OURS = saturated (8-epoch) per-domain peaks. +DOMS = ["cw", "fqa", "general", "code", "math"] +NAIVE = [2.591, 2.775, 3.000, 3.211, 4.632] # naive same-data CorDA-MoS from D0 (per-domain peak) — loses gen +GEN = [2.647, 2.814, 3.049, 3.204, 4.698] # generalist, single best-avg checkpoint +MOS = [2.721, 2.907, 3.108, 3.338, 4.922] # CorDA-MoS warm-from-gen, SATURATED per-domain peak — OURS + +x = np.arange(len(DOMS)); w = 0.27 +fig, ax = plt.subplots(figsize=(10.5, 5.4)) +b0 = ax.bar(x - w, NAIVE, w, label="naive same-data (from D0, before warm-start) — loses", color="#c0392b") +b1 = ax.bar(x, GEN, w, label="generalist (same 250k data)", color="#9aa7b8") +b2 = ax.bar(x + w, MOS, w, label="CorDA-MoS warm-from-gen (OURS, same data) — wins", color="#1b5e20") +for bars in (b0, b1, b2): + for b in bars: + ax.text(b.get_x()+b.get_width()/2, b.get_height()+0.02, f"{b.get_height():.2f}", ha="center", va="bottom", fontsize=7.5) +# show the advance our method makes over the naive version +for j in range(len(DOMS)): + gain = MOS[j] - NAIVE[j] + ax.text(x[j]+w, MOS[j]+0.20, f"+{gain:.2f} vs naive", ha="center", fontsize=7, color="#1b5e20", fontweight="bold") +ax.set_xticks(x); ax.set_xticklabels(DOMS) +ax.set_ylabel("held-out accept length (AL)") +ax.set_ylim(2.0, 5.4) +ax.set_title("Method > Data (same 250k): naive same-data split (from D0) LOSES to gen;\n" + "our warm-from-gen CorDA-MoS BEATS gen on all 5 domains — the gap shows the method's contribution", fontsize=10.5) +ax.legend(loc="upper left", fontsize=8.5); ax.grid(axis="y", ls=":", alpha=0.4) +ax.text(0.58, 0.74, f"avg AL: naive {np.mean(NAIVE):.3f} < gen {np.mean(GEN):.3f} < ours {np.mean(MOS):.3f}", + transform=ax.transAxes, fontsize=9.5, va="top", ha="center", + bbox=dict(boxstyle="round,pad=0.3", fc="#e8f3e8", ec="#1b5e20")) +fig.tight_layout() +fig.savefig("fig_exp5_method_vs_data.png", dpi=140, bbox_inches="tight") +print("OK wrote /tmp/fig_exp5_method_vs_data.png") diff --git a/recipes/plotting/v1/fig_fusion_combined.py b/recipes/plotting/v1/fig_fusion_combined.py new file mode 100644 index 0000000000000000000000000000000000000000..8cb08b1c8e8e77a1543084ced6d12f0b99b5db37 --- /dev/null +++ b/recipes/plotting/v1/fig_fusion_combined.py @@ -0,0 +1,31 @@ +#!/usr/bin/env python3 +# Combined fusion figure (replaces separate Exp3 + Exp6 figs): +# per domain, three bars — generalist (baseline) -> small-data fusion (CorDA-MoS warm, SAME data as gen) +# -> big-data fusion (merged 5 big-data specialists). Shows the fusion ladder: method gain (same data), +# then method+data gain. +import matplotlib; matplotlib.use("Agg") +import matplotlib.pyplot as plt +import numpy as np +DOMS = ["cw", "fqa", "general", "code", "math"] # ascending -> math tallest at right +GEN = [2.647, 2.814, 3.049, 3.204, 4.698] # generalist (single ckpt) avg 3.282 +SMALLFUS = [2.721, 2.907, 3.108, 3.338, 4.922] # CorDA-MoS warm, SAME 250k data avg 3.399 +BIGFUS = [2.857, 3.061, 3.227, 3.392, 5.247] # merged big-data drafter (250k/dom) avg 3.557 +x = np.arange(len(DOMS)); w = 0.27 +fig, ax = plt.subplots(figsize=(11, 5.6)) +b0 = ax.bar(x - w, GEN, w, label=f"generalist (baseline) avg {np.mean(GEN):.3f}", color="#9aa7b8") +b1 = ax.bar(x, SMALLFUS, w, label=f"small-data fusion · CorDA-MoS (SAME data as gen) avg {np.mean(SMALLFUS):.3f}", color="#2a7fb8") +b2 = ax.bar(x + w, BIGFUS, w, label=f"big-data fusion · merged drafter (250k/domain) avg {np.mean(BIGFUS):.3f}", color="#1b5e20") +for bars in (b0, b1, b2): + for b in bars: + ax.text(b.get_x()+b.get_width()/2, b.get_height()+0.02, f"{b.get_height():.2f}", ha="center", va="bottom", fontsize=7.5) +ax.set_xticks(x); ax.set_xticklabels(DOMS) +ax.set_ylabel("held-out accept length (AL)"); ax.set_ylim(2.2, 5.6) +ax.set_title("Fusion ladder: same-data fusion already beats the generalist on all 5 domains (method),\n" + "more data per domain lifts it further — both share ONE attention (inference cost = one drafter)", fontsize=10.5) +ax.legend(loc="upper left", fontsize=8.6); ax.grid(axis="y", ls=":", alpha=0.4) +ax.text(0.60, 0.70, f"avg AL: gen {np.mean(GEN):.3f} < small-data fusion {np.mean(SMALLFUS):.3f} < big-data fusion {np.mean(BIGFUS):.3f}", + transform=ax.transAxes, fontsize=9.5, va="top", ha="center", + bbox=dict(boxstyle="round,pad=0.3", fc="#e8f3e8", ec="#1b5e20")) +fig.tight_layout() +fig.savefig("fig_fusion_combined.png", dpi=140, bbox_inches="tight") +print("OK wrote fig_fusion_combined.png") diff --git a/recipes/plotting/v1/fig_matrix_and_arms.py b/recipes/plotting/v1/fig_matrix_and_arms.py new file mode 100644 index 0000000000000000000000000000000000000000..51bd2a39acbb3f22e127eca52e79e0167030c6b6 --- /dev/null +++ b/recipes/plotting/v1/fig_matrix_and_arms.py @@ -0,0 +1,118 @@ +#!/usr/bin/env python3 +# Forgetting matrix + per-arm epoch-matched comparisons (reasonmix DFlash). +# All numbers baked from same-protocol bench (DFLASH 8,1,1,16, ROUTED=0, thinking-on, reasonmix held-out). +import matplotlib; matplotlib.use("Agg") +import matplotlib.pyplot as plt +from matplotlib.colors import LinearSegmentedColormap +from matplotlib.patches import Rectangle +import numpy as np +import os + +OUT = os.environ.get("MOS_FIG_OUT", ".") +DOM = ["code", "math", "factual_qa", "creative_writing", "general"] +SHORT = ["code", "math", "fqa", "cw", "general"] + +# ---------- FIG 1: 5x5 forgetting matrix (same-protocol) ---------- +SPECS = ["code", "math", "factual_qa", "creative_writing", "general"] +M = np.array([ + [3.340, 4.094, 2.612, 2.419, 2.847], # code-spec + [2.853, 5.190, 2.501, 2.296, 2.842], # math-spec + [2.914, 4.278, 2.902, 2.491, 3.026], # factual_qa-spec + [2.805, 3.892, 2.707, 2.861, 2.948], # cw-spec + [2.992, 4.614, 2.773, 2.627, 3.117], # general-spec +]) +GEN = np.array([3.209, 4.698, 2.814, 2.636, 3.049]) # generalist (same protocol, ep3) +D = M - GEN[None, :] # delta vs generalist, per column + +matrix_short = ["Code", "Math", "Fact.", "Creat.", "Gen."] +paper_fs = 9.6 # remains at least 9 pt after final single-column placement +paper_diverging = LinearSegmentedColormap.from_list( + "paper_diverging", ["#c6dbef", "#ffffff", "#fdd0a2"] +) +fig, ax = plt.subplots(figsize=(3.3, 3.0)) +vmax = np.abs(D).max() +im = ax.imshow(D, cmap=paper_diverging, vmin=-vmax, vmax=vmax, aspect="auto") +for i in range(5): + for j in range(5): + delta = f"{D[i,j]:+.2f}".replace("+0.", "+.").replace("-0.", "-.") + ax.text(j, i, f"{M[i,j]:.2f}\n{delta}", ha="center", va="center", + fontsize=paper_fs, linespacing=0.95, + fontweight=("bold" if i == j else "normal")) + if i == j: + ax.add_patch(Rectangle((j - 0.46, i - 0.46), 0.92, 0.92, + fill=False, edgecolor="black", linewidth=1.0)) +ax.set_xticks(range(5)); ax.set_xticklabels(matrix_short, fontsize=paper_fs) +ax.set_yticks(range(5)); ax.set_yticklabels(matrix_short, fontsize=paper_fs) +ax.set_xlabel("Evaluation domain", fontsize=paper_fs, labelpad=3) +ax.set_ylabel("Specialist", fontsize=paper_fs, labelpad=3) +ax.tick_params(axis="both", labelsize=paper_fs, width=0.6, length=2.5) +for spine in ax.spines.values(): + spine.set_linewidth(0.6) +cb = fig.colorbar(im, ax=ax, fraction=0.050, pad=0.025) +cb.set_label(r"$\Delta$ AL vs. generalist", fontsize=paper_fs, labelpad=3) +cb.ax.tick_params(labelsize=paper_fs, width=0.6, length=2.5) +cb.outline.set_linewidth(0.6) +fig.tight_layout(pad=0.25) +fig.savefig(f"{OUT}/fig_forgetting_matrix.png", dpi=300, bbox_inches="tight") +plt.close(fig) + +# ---------- generic per-arm grouped-bar (epoch-matched: arm@best vs gen@same epoch) ---------- +def arm_vs_gen(fname, title, spec, gen, ep_lbl, spec_name, spec_color): + x = np.arange(5); w = 0.38 + fig, ax = plt.subplots(figsize=(8.4, 4.8)) + b1 = ax.bar(x - w/2, gen, w, label="Generalist (same epoch)", color="#9aa7b8") + b2 = ax.bar(x + w/2, spec, w, label=spec_name, color=spec_color) + for bars in (b1, b2): + for b in bars: + ax.text(b.get_x()+b.get_width()/2, b.get_height()+0.02, f"{b.get_height():.2f}", + ha="center", va="bottom", fontsize=8) + for j in range(5): + d = spec[j] - gen[j] + ax.text(x[j], max(spec[j], gen[j]) + 0.16, f"{d:+.3f}", ha="center", fontsize=9, + fontweight="bold", color=("#2e7d32" if d > 0 else "#c62828")) + ax.text(x[j], min(spec[j], gen[j]) - 0.001, ep_lbl[j], ha="center", va="top", fontsize=7, color="#555") + ax.set_xticks(x); ax.set_xticklabels(SHORT) + ax.set_ylabel("held-out accept length (AL)") + ax.set_ylim(2.0, max(spec.max(), gen.max()) + 0.5) + ax.set_title(title, fontsize=11) + ax.legend(loc="upper left", fontsize=9); ax.grid(axis="y", ls=":", alpha=0.4) + fig.tight_layout(); fig.savefig(f"{OUT}/{fname}", dpi=140, bbox_inches="tight"); plt.close(fig) + +# FIG 2: big-data specialist (250k, full params, from D0) — best epoch vs gen@same epoch +bd_spec = np.array([3.340, 5.190, 2.932, 2.861, 3.117]); bd_gen = np.array([3.205, 4.698, 2.767, 2.651, 3.075]) +bd_ep = ["ep2", "ep3", "ep2", "ep3", "ep3"] +arm_vs_gen("fig_bigdata_vs_gen.png", + "Big-data specialist (250k single-domain, full params, from D0)\nspecialist@best-epoch vs Generalist@same-epoch", + bd_spec, bd_gen, bd_ep, "Big-data specialist", "#2e7d32") + +# FIG 3: small-data specialist (oracle: baseline volume, full params, from D0) — SEPARATE +sd_spec = np.array([3.162, 4.493, 2.683, 2.536, 2.908]); sd_gen = np.array([3.205, 4.694, 2.814, 2.596, 3.075]) +sd_ep = ["ep2", "ep2", "ep3", "ep1", "ep3"] +arm_vs_gen("fig_smalldata_vs_gen.png", + "Small-data specialist (baseline volume, full params, from D0)\nspecialist@best-epoch vs Generalist@same-epoch (loses on every domain)", + sd_spec, sd_gen, sd_ep, "Small-data specialist", "#607d8b") + +# ---------- FIG 4: warm specialist (cleanA: MLP-only, frozen backbone, from generalist) saturation ---------- +warm = { + "code": ([700, 1400, 2100, 2800, 2840], [3.210, 3.214, 3.208, 3.215, 3.198]), + "math": ([250, 500, 750, 984], [4.748, 4.690, 4.681, 4.675]), + "factual_qa": ([350, 700, 1050, 1390], [2.789, 2.843, 2.790, 2.861]), + "creative_writing": ([230, 460, 690, 913], [2.670, 2.654, 2.646, 2.671]), + "general": ([420, 840, 1260, 1674], [3.106, 3.093, 3.079, 3.058]), +} +warm_gen = {"code": 3.205, "math": 4.698, "factual_qa": 2.814, "creative_writing": 2.651, "general": 3.086} +fig, axes = plt.subplots(1, 5, figsize=(15, 3.4)) +for ax, d, s in zip(axes, DOM, SHORT): + steps, al = warm[d]; frac = np.array(steps) / steps[-1] + ax.plot(frac, al, "o-", color="#b8860b", lw=1.8, label="warm spec (MLP-only)") + ax.axhline(warm_gen[d], ls="--", color="#1565c0", lw=1.3, label="Generalist (start)") + ax.set_title(s, fontsize=10); ax.set_xlabel("frac of 1 epoch"); ax.grid(alpha=0.3) + ax.set_xlim(0, 1.02) + lo = min(al + [warm_gen[d]]); hi = max(al + [warm_gen[d]]) + ax.set_ylim(lo - 0.04, hi + 0.04) +axes[0].set_ylabel("held-out AL"); axes[0].legend(fontsize=7, loc="lower right") +fig.suptitle("Warm spec (continue Generalist's MLP only, backbone frozen): AL stays within ~±0.04 of Generalist " + "— flat / declining (math, general), peaks early. MLP already saturated → can't push AL up.", fontsize=10.5, y=1.04) +fig.tight_layout(); fig.savefig(f"{OUT}/fig_warm_saturation.png", dpi=140, bbox_inches="tight"); plt.close(fig) + +print("OK wrote: fig_forgetting_matrix.png fig_bigdata_vs_gen.png fig_smalldata_vs_gen.png fig_warm_saturation.png") diff --git a/recipes/plotting/v1/fig_merged_vs_specialist.py b/recipes/plotting/v1/fig_merged_vs_specialist.py new file mode 100644 index 0000000000000000000000000000000000000000..9777ce7d8bd8c138ea881c70ae073707b9ab0fb5 --- /dev/null +++ b/recipes/plotting/v1/fig_merged_vs_specialist.py @@ -0,0 +1,39 @@ +#!/usr/bin/env python3 +# Headline: the merged drafter (1 shared CorDA-fused attention + per-domain MLP) vs 5 separate specialists vs generalist. +import matplotlib; matplotlib.use("Agg") +import matplotlib.pyplot as plt +import numpy as np + +# ordered by merged AL ascending -> tallest (math) at the right end +DOMS = ["creative_writing", "factual_qa", "general", "code", "math"] +SHORT = ["cw", "fqa", "general", "code", "math"] +GEN = [2.636, 2.814, 3.049, 3.209, 4.698] +MERGED = [2.857, 3.061, 3.227, 3.392, 5.247] # all 5 domains: MLP retrained on the fused attention +SPEC = [2.861, 2.902, 3.117, 3.340, 5.190] +RETRAINED = [True, True, True, True, True] # all retrained -> all beat their specialist + +x = np.arange(len(DOMS)); w = 0.26 +fig, ax = plt.subplots(figsize=(9.5, 5.0)) +b1 = ax.bar(x - w, GEN, w, label="Generalist (monolithic)", color="#9aa7b8") +b2 = ax.bar(x, MERGED, w, label="Merged drafter (1 shared attn + per-domain MLP)", color="#2e7d32") +b3 = ax.bar(x + w, SPEC, w, label="5 separate specialists", color="#b8860b") +for bars in (b1, b2, b3): + for b in bars: + ax.text(b.get_x()+b.get_width()/2, b.get_height()+0.02, f"{b.get_height():.2f}", + ha="center", va="bottom", fontsize=7.5) +for j in range(len(DOMS)): + if RETRAINED[j]: + ax.text(x[j], MERGED[j]+0.18, "retrained", ha="center", fontsize=6.5, color="#2e7d32") +ax.set_xticks(x); ax.set_xticklabels(SHORT) +ax.set_ylabel("held-out accept length (AL)") +ax.set_ylim(2.0, 5.6) +ax.set_title("Merged drafter > 5 separate specialists (avg 3.557 vs 3.482) with ONE shared attention, >> generalist (3.281)\n" + "merged = CorDA-fused shared attention + per-domain MLP (all 5 retrained on the fused attn; every domain beats its specialist)", fontsize=10.5) +ax.legend(loc="upper left", fontsize=9); ax.grid(axis="y", ls=":", alpha=0.4) +# avg annotation +ax.text(0.46, 0.82, f"avg AL: merged {np.mean(MERGED):.3f} | specialists {np.mean(SPEC):.3f} | gen {np.mean(GEN):.3f}", + transform=ax.transAxes, fontsize=9, va="top", + bbox=dict(boxstyle="round,pad=0.3", fc="#eef7ee", ec="#2e7d32")) +fig.tight_layout() +fig.savefig("fig_merged_vs_specialist.png", dpi=140, bbox_inches="tight") +print("OK wrote /tmp/fig_merged_vs_specialist.png") diff --git a/recipes/plotting/v1/fig_serving_specialist.py b/recipes/plotting/v1/fig_serving_specialist.py new file mode 100644 index 0000000000000000000000000000000000000000..a7f250b09729a2c6eac0652926484385338906ae --- /dev/null +++ b/recipes/plotting/v1/fig_serving_specialist.py @@ -0,0 +1,58 @@ +import matplotlib; matplotlib.use("Agg") +import matplotlib.pyplot as plt + +# ---- fig_serving: (a) single-stream tokens/s, (b) garbage tau tradeoff ---- +fig, (a, b) = plt.subplots(1, 2, figsize=(12.6, 4.4), gridspec_kw={"width_ratios": [1, 1.15]}) +for ax in (a, b): + for s in ["top", "right"]: ax.spines[s].set_visible(False) + for s in ["left", "bottom"]: ax.spines[s].set_color("#c9ced6") + ax.tick_params(colors="#444", labelsize=10.5) + +names = ["target only", "target +\ngeneralist drafter", "MoS, real router\n(5 experts + router)"] +tps = [184, 391, 415] +cols = ["#8a8f98", "#4DABF7", "#9C36B5"] +bars = a.barh(names, tps, color=cols, height=0.62) +a.invert_yaxis() +for r, v, al in zip(bars, tps, ["", "AL 3.40", "AL 3.65 · acc 0.878"]): + a.text(v + 6, r.get_y() + r.get_height() / 2, f"{v}", va="center", fontsize=12, fontweight="bold", color=r.get_facecolor()) + if al: + a.text(8, r.get_y() + r.get_height() / 2, al, va="center", fontsize=9.5, color="white", fontweight="bold") +a.set_xlim(0, 500) +a.set_xlabel("tokens/s (single stream, H200, thinking on)", fontsize=11) +a.set_title("Serving speed with the real router", fontsize=12, fontweight="bold", loc="left") + +taus = [0.2, 0.3, 0.4, 0.5, 0.6, 0.7] +junk = [60.0, 66.7, 68.9, 73.3, 73.3, 75.6] +real = [3.5, 5.5, 6.0, 6.0, 6.5, 9.5] +b.axvspan(0.4, 0.5, color="#E9FAC8", alpha=0.7, zorder=0) +b.plot(taus, junk, "o-", color="#2F9E44", lw=2.2, ms=5, label="junk requests sent to garbage (%)") +b.plot(taus, real, "s-", color="#E8590C", lw=2.2, ms=5, label="real-domain requests mis-sent (%)") +for t, j in zip(taus, junk): b.text(t, j + 2.5, f"{j:.0f}", ha="center", fontsize=9, color="#2F9E44") +for t, r_ in zip(taus, real): b.text(t, r_ + 2.5, f"{r_:.0f}", ha="center", fontsize=9, color="#E8590C") +b.text(0.45, 96, "τ = 0.4–0.5", ha="center", fontsize=10, color="#5c940d", fontweight="bold") +b.set_ylim(0, 105); b.set_xlim(0.17, 0.73) +b.set_xlabel("garbage threshold τ (raw 4-domain max prob)", fontsize=11) +b.set_ylabel("% of requests", fontsize=11) +b.set_title("Garbage fallback v1 (no training): expected AL unchanged", fontsize=12, fontweight="bold", loc="left") +b.legend(frameon=False, fontsize=9.5, loc="center right") +plt.tight_layout() +plt.savefig("fig_serving.png", dpi=150, bbox_inches="tight") + +# ---- fig_specialist_overall: weighted overall bars ---- +fig2, c = plt.subplots(figsize=(8.6, 4.2)) +for s in ["top", "right"]: c.spines[s].set_visible(False) +for s in ["left", "bottom"]: c.spines[s].set_color("#c9ced6") +c.tick_params(colors="#444", labelsize=10.5) +labels = ["generalist", "code specialist\n(from generalist)", "code specialist\n(from scratch)", "MoS\n(from D0)", "MoS\n(from generalist)"] +vals = [3.39, 3.40, 3.18, 3.59, 3.65] +ccols = ["#8a8f98", "#C92A2A", "#5F3DC4", "#2F9E44", "#9C36B5"] +bars = c.bar(labels, vals, color=ccols, width=0.62) +for r, v in zip(bars, vals): + c.text(r.get_x() + r.get_width() / 2, v + 0.012, f"{v:.2f}", ha="center", fontsize=12, fontweight="bold", color=r.get_facecolor()) +c.axhline(3.39, ls=(0, (3, 3)), lw=1.1, color="#8a8f98", alpha=0.8) +c.set_ylim(3.0, 3.78) +c.set_ylabel("AL, weighted by traffic share", fontsize=11) +c.set_title("One specialist ≈ generalist at best; MoS beats both (same active params)", fontsize=12, fontweight="bold", loc="left") +plt.tight_layout() +plt.savefig("fig_specialist_overall.png", dpi=150, bbox_inches="tight") +print("saved both") diff --git a/recipes/plotting/v1/plot_main_results.py b/recipes/plotting/v1/plot_main_results.py new file mode 100644 index 0000000000000000000000000000000000000000..dd5cb519f44258d2c7c4800d87d681af66e5e3bd --- /dev/null +++ b/recipes/plotting/v1/plot_main_results.py @@ -0,0 +1,380 @@ +#!/usr/bin/env python3 +"""Render the three-panel MoS main-results figure from frozen evidence. + +All three panels use the fixed-seed, five-domain R1 evaluation. Panel (a) +compares the matched-domain profiles of the Generalist and the two MoS +initializations; panels (b)--(c) show the corresponding MoS routing matrices. +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +import matplotlib as mpl +import matplotlib.pyplot as plt +import numpy as np +from matplotlib.colors import LinearSegmentedColormap, Normalize +from matplotlib.patches import Rectangle + + +REPO_ROOT = Path(__file__).resolve().parents[3] +DEFAULT_EVIDENCE = ( + REPO_ROOT + / "paper" + / "submission" + / "evidence" + / "r1_mainfig_seed20260719_20260721T0602Z_cells_summary.json" +) +DEFAULT_OUTPUT = ( + REPO_ROOT / "paper" / "submission" / "figures" / "fig_main_results.png" +) + +DOMAINS = ["code", "math", "factual_qa", "creative_writing", "general"] +DOMAIN_LABELS = ["Code", "Math", "Factual QA", "Creative", "General"] + +# Restrained, color-blind-safe palette. Shape and line style also distinguish +# methods, so the figure remains legible in grayscale. +INK = "#25313B" +MUTED = "#68747E" +GRID = "#E2E7EA" +GENERALIST = "#7F8790" +D0 = "#2F9E44" +WARM = "#9C36B5" +LIGHT_RULE = "#C8D0D5" + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--evidence", type=Path, default=DEFAULT_EVIDENCE) + parser.add_argument("--output", type=Path, default=DEFAULT_OUTPUT) + return parser.parse_args() + + +def configure_style() -> None: + mpl.rcParams.update( + { + "font.family": "sans-serif", + "font.sans-serif": [ + "Arial", + "Helvetica", + "Liberation Sans", + "DejaVu Sans", + ], + "font.size": 8.0, + "axes.titlesize": 9.3, + "axes.labelsize": 8.3, + "xtick.labelsize": 7.4, + "ytick.labelsize": 7.4, + "legend.fontsize": 7.3, + "pdf.fonttype": 42, + "ps.fonttype": 42, + "axes.linewidth": 0.65, + "savefig.bbox": "tight", + "savefig.pad_inches": 0.035, + } + ) + + +def matrix_from_evidence(evidence: dict, key: str) -> np.ndarray: + mapping = evidence[key] + matrix = np.asarray( + [[float(mapping[row][column]) for column in DOMAINS] for row in DOMAINS], + dtype=float, + ) + for column in range(len(DOMAINS)): + if int(np.argmax(matrix[:, column])) != column: + raise ValueError(f"{key}: matched MLP is not best in column {DOMAINS[column]}") + return matrix + + +def generalist_from_evidence(evidence: dict) -> np.ndarray: + mapping = evidence["panel_d_generalist"] + return np.asarray([float(mapping[domain]) for domain in DOMAINS], dtype=float) + + +def panel_title(ax: plt.Axes, letter: str, title: str) -> None: + # Use a point-based offset so the letter-to-title gap is physically + # identical in the full-width trajectory and the half-width matrices. + origin = (-0.055, 1.075) + ax.text( + *origin, + letter, + transform=ax.transAxes, + ha="left", + va="bottom", + fontsize=9.8, + fontweight="bold", + color=INK, + ) + ax.annotate( + title, + xy=origin, + xycoords="axes fraction", + xytext=(18, 0), + textcoords="offset points", + ha="left", + va="bottom", + fontsize=8.6, + fontweight="bold", + color=INK, + ) + + +def quiet_axes(ax: plt.Axes) -> None: + ax.spines["top"].set_visible(False) + ax.spines["right"].set_visible(False) + ax.spines["left"].set_color(LIGHT_RULE) + ax.spines["bottom"].set_color(LIGHT_RULE) + ax.tick_params(color=LIGHT_RULE, labelcolor=INK, width=0.65, length=2.8) + + +def draw_profile_panel( + ax: plt.Axes, + generalist: np.ndarray, + d0_diag: np.ndarray, + warm_diag: np.ndarray, +) -> None: + panel_title(ax, "A", "Matched-domain acceptance across five domains") + quiet_axes(ax) + ax.grid(axis="y", color=GRID, lw=0.6, alpha=0.9, zorder=0) + x = np.arange(len(DOMAINS)) + ax.plot( + x, + generalist, + color=GENERALIST, + lw=1.65, + ls="-", + marker="o", + ms=4.2, + mfc="white", + mec=GENERALIST, + mew=1.0, + alpha=0.90, + label="Generalist", + zorder=3, + ) + ax.plot( + x, + d0_diag, + color=D0, + lw=1.75, + marker="o", + ms=4.3, + mfc="white", + mec=D0, + mew=1.0, + label="D0-init MoS", + zorder=5, + ) + ax.plot( + x, + warm_diag, + color=WARM, + lw=1.75, + marker="s", + ms=4.2, + mfc="white", + mec=WARM, + mew=1.0, + label="G-init MoS", + zorder=6, + ) + + for index, value in enumerate(generalist): + ax.annotate( + f"{value:.3f}", + xy=(x[index], value), + xytext=(0, -7), + textcoords="offset points", + ha="center", + va="top", + fontsize=5.7, + color=MUTED, + ) + for index, value in enumerate(warm_diag): + ax.annotate( + f"{value:.3f}", + xy=(x[index], value), + xytext=(0, 6), + textcoords="offset points", + ha="center", + va="bottom", + fontsize=5.8, + color=INK, + fontweight="bold", + ) + + ax.set_xlim(-0.35, len(DOMAINS) - 0.65) + ax.set_ylim(2.55, 5.78) + ax.set_xticks(x, labels=DOMAIN_LABELS) + ax.set_yticks([2.8, 3.4, 4.0, 4.6, 5.2, 5.8]) + ax.set_ylabel("Acceptance length") + ax.legend( + loc="upper right", + bbox_to_anchor=(1.0, 1.02), + frameon=False, + ncol=3, + handlelength=2.4, + borderaxespad=0.1, + columnspacing=1.15, + handletextpad=0.4, + labelspacing=0.25, + fontsize=5.8, + ) + + +HEATMAP_D0_CMAP = LinearSegmentedColormap.from_list( + "d0_matched_regret", + ["#F7FAF7", "#DDEFE1", "#A7D7B1", "#68B97A", D0], +) +HEATMAP_WARM_CMAP = LinearSegmentedColormap.from_list( + "warm_matched_regret", + ["#FBF8FC", "#F0E0F4", "#D9B7E2", "#BC79CB", WARM], +) +HEATMAP_NORM = Normalize(vmin=-1.30, vmax=0.0) + + +def draw_matrix_panel( + ax: plt.Axes, + matrix: np.ndarray, + letter: str, + title: str, + cmap: LinearSegmentedColormap, + diagonal_edge: str, +) -> mpl.image.AxesImage: + regret = matrix - np.diag(matrix)[None, :] + image = ax.imshow(regret, cmap=cmap, norm=HEATMAP_NORM, aspect="equal") + panel_title(ax, letter, title) + ax.set_xticks(range(len(DOMAINS)), labels=DOMAIN_LABELS) + ax.set_yticks(range(len(DOMAINS)), labels=DOMAIN_LABELS) + ax.tick_params(axis="x", rotation=29, length=0, pad=2.2, labelsize=6.2) + ax.tick_params(axis="y", length=0, pad=2.6, labelsize=6.5) + ax.set_ylabel("Selected MLP", labelpad=2.5, fontsize=7.0) + + for row in range(len(DOMAINS)): + for column in range(len(DOMAINS)): + value = matrix[row, column] + normalized = HEATMAP_NORM(regret[row, column]) + text_color = "white" if normalized > 0.68 else INK + ax.text( + column, + row, + f"{value:.3f}", + ha="center", + va="center", + fontsize=6.0, + color=text_color, + fontweight="bold" if row == column else "normal", + ) + if row == column: + ax.add_patch( + Rectangle( + (column - 0.48, row - 0.48), + 0.96, + 0.96, + facecolor="none", + edgecolor=diagonal_edge, + linewidth=1.45, + ) + ) + + ax.set_xticks(np.arange(-0.5, len(DOMAINS), 1), minor=True) + ax.set_yticks(np.arange(-0.5, len(DOMAINS), 1), minor=True) + ax.grid(which="minor", color="white", linestyle="-", linewidth=1.15) + ax.tick_params(which="minor", bottom=False, left=False) + for spine in ax.spines.values(): + spine.set_visible(False) + return image + + +def main() -> None: + args = parse_args() + configure_style() + evidence = json.loads(args.evidence.read_text()) + if not evidence.get("passed"): + raise ValueError("R1 evidence is not marked passed") + if evidence.get("cells_total") != 52 or evidence.get("cells_passed") != 52: + raise ValueError("R1 evidence is not complete (expected 52/52 cells)") + + d0_matrix = matrix_from_evidence(evidence, "panel_b_matrix_dflash_init") + warm_matrix = matrix_from_evidence(evidence, "panel_c_matrix_warm_start") + generalist = generalist_from_evidence(evidence) + + if not np.all(np.diag(d0_matrix) > generalist): + raise ValueError("DFlash-initialized MoS is not above Generalist in every domain") + if not np.all(np.diag(warm_matrix) > generalist): + raise ValueError("warm-started MoS is not above Generalist in every domain") + + # Match the intended AAAI double-column physical width. Raising DPI, rather + # than drawing an oversized canvas and shrinking it in LaTeX, preserves the + # configured 7--10 pt typography at publication size. + fig = plt.figure(figsize=(7.15, 5.00), facecolor="white") + outer = fig.add_gridspec( + 2, + 1, + height_ratios=[0.82, 1.18], + hspace=0.46, + left=0.075, + right=0.985, + top=0.945, + bottom=0.180, + ) + ax_a = fig.add_subplot(outer[0, 0]) + matrices = outer[1, 0].subgridspec(1, 2, wspace=0.28) + ax_b = fig.add_subplot(matrices[0, 0]) + ax_c = fig.add_subplot(matrices[0, 1]) + + draw_profile_panel( + ax_a, + generalist, + np.diag(d0_matrix), + np.diag(warm_matrix), + ) + d0_image = draw_matrix_panel( + ax_b, + d0_matrix, + "B", + "DFlash-initialized MoS", + HEATMAP_D0_CMAP, + "#226F32", + ) + warm_image = draw_matrix_panel( + ax_c, + warm_matrix, + "C", + "Generalist-warm-started MoS", + HEATMAP_WARM_CMAP, + "#6F277D", + ) + + # Separate color strips preserve the original green/purple recipe identity + # while keeping an identical quantitative scale in both matrices. + for image, position in ( + (d0_image, [0.145, 0.065, 0.29, 0.010]), + (warm_image, [0.575, 0.065, 0.29, 0.010]), + ): + cbar_ax = fig.add_axes(position) + cbar = fig.colorbar(image, cax=cbar_ax, orientation="horizontal") + cbar.set_ticks([-1.2, -0.6, 0.0]) + cbar.ax.tick_params(labelsize=5.7, length=1.8, color=LIGHT_RULE) + cbar.outline.set_visible(False) + fig.text( + 0.505, + 0.014, + r"Shade: column-wise $\Delta$AL from the matched MLP", + ha="center", + va="bottom", + fontsize=6.0, + color=INK, + ) + + args.output.parent.mkdir(parents=True, exist_ok=True) + fig.savefig(args.output, dpi=420, facecolor="white") + plt.close(fig) + print(f"saved {args.output}") + + +if __name__ == "__main__": + main() diff --git a/recipes/plotting/v1/plot_mos_5x5_gains.py b/recipes/plotting/v1/plot_mos_5x5_gains.py new file mode 100644 index 0000000000000000000000000000000000000000..412c22743276f1d6dd86356bac791e069cf068b7 --- /dev/null +++ b/recipes/plotting/v1/plot_mos_5x5_gains.py @@ -0,0 +1,929 @@ +#!/usr/bin/env python3 +"""Render separate 5x5 MoS routing matrices with Generalist gains. + +The frozen R1 evidence contains two selected-checkpoint matrices: + +1. MoS initialized from the public DFlash drafter (D0-init). +2. MoS warm-started from the trained Generalist (G-init). + +The first two figures show all selected-MLP x evaluation-domain AL cells. The +right panel reports the matched-domain diagonal's absolute and relative +improvement over one fixed-seed evaluation of the selected Generalist +checkpoint. A third figure shows that Generalist baseline across the five +evaluation domains. +""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +import matplotlib as mpl +import matplotlib.pyplot as plt +import numpy as np +from matplotlib.colors import LinearSegmentedColormap, Normalize +from matplotlib.patches import Rectangle + + +REPO_ROOT = Path(__file__).resolve().parents[3] +DEFAULT_EVIDENCE = ( + REPO_ROOT + / "paper" + / "submission" + / "evidence" + / "r1_mainfig_seed20260719_20260721T0602Z_cells_summary.json" +) +DEFAULT_OUTPUT_DIR = REPO_ROOT / "paper" / "submission" / "figures" + +DOMAINS = ["code", "math", "factual_qa", "creative_writing", "general"] +DOMAIN_LABELS = ["Code", "Math", "Factual QA", "Creative", "General"] + +INK = "#25313B" +MUTED = "#68747E" +RULE = "#D9DFE3" +ROW_FILL = "#F3F5F6" +D0 = "#2F9E44" +WARM = "#9C36B5" +GENERALIST = "#7F8790" + +HEATMAP_NORM = Normalize(vmin=-1.30, vmax=0.0) +D0_CMAP = LinearSegmentedColormap.from_list( + "d0_regret", ["#F7FAF7", "#DDEFE1", "#A7D7B1", "#68B97A", D0] +) +WARM_CMAP = LinearSegmentedColormap.from_list( + "ginit_regret", ["#FBF8FC", "#F0E0F4", "#D9B7E2", "#BC79CB", WARM] +) +ABSOLUTE_CMAP = LinearSegmentedColormap.from_list( + "absolute_al", ["#F2F7FB", "#C9DEEE", "#80B7D5", "#3182BD", "#12538A"] +) +ABSOLUTE_NORM = Normalize(vmin=2.25, vmax=5.60) + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--evidence", type=Path, default=DEFAULT_EVIDENCE) + parser.add_argument("--output-dir", type=Path, default=DEFAULT_OUTPUT_DIR) + return parser.parse_args() + + +def configure_style() -> None: + mpl.rcParams.update( + { + "font.family": "sans-serif", + "font.sans-serif": [ + "Arial", + "Helvetica", + "Liberation Sans", + "DejaVu Sans", + ], + "font.size": 8.0, + "axes.titlesize": 10.0, + "axes.labelsize": 8.2, + "xtick.labelsize": 7.4, + "ytick.labelsize": 7.4, + "pdf.fonttype": 42, + "ps.fonttype": 42, + "savefig.bbox": "tight", + "savefig.pad_inches": 0.035, + } + ) + + +def load_evidence(path: Path) -> tuple[dict, np.ndarray]: + evidence = json.loads(path.read_text()) + if not evidence.get("passed"): + raise ValueError("R1 evidence is not marked passed") + if evidence.get("cells_total") != 52 or evidence.get("cells_passed") != 52: + raise ValueError("R1 evidence is incomplete; expected 52/52 passed cells") + generalist = np.asarray( + [float(evidence["panel_d_generalist"][domain]) for domain in DOMAINS], + dtype=float, + ) + return evidence, generalist + + +def matrix_from_evidence(evidence: dict, key: str) -> np.ndarray: + mapping = evidence[key] + matrix = np.asarray( + [[float(mapping[row][column]) for column in DOMAINS] for row in DOMAINS], + dtype=float, + ) + for column, domain in enumerate(DOMAINS): + if int(np.argmax(matrix[:, column])) != column: + raise ValueError(f"{key}: matched MLP is not best for {domain}") + return matrix + + +def draw_matrix( + ax: plt.Axes, + matrix: np.ndarray, + cmap: LinearSegmentedColormap, + accent: str, +) -> mpl.image.AxesImage: + regret = matrix - np.diag(matrix)[None, :] + image = ax.imshow(regret, cmap=cmap, norm=HEATMAP_NORM, aspect="equal") + ax.set_xticks(range(5), labels=DOMAIN_LABELS) + ax.set_yticks(range(5), labels=DOMAIN_LABELS) + ax.tick_params(axis="x", rotation=28, length=0, pad=3.0) + ax.tick_params(axis="y", length=0, pad=3.0) + ax.xaxis.set_label_position("top") + ax.set_xlabel("Evaluation domain", labelpad=8.5, fontweight="bold") + ax.set_ylabel("Selected MLP", labelpad=6.0, fontweight="bold") + + for row in range(5): + for column in range(5): + value = matrix[row, column] + normalized = HEATMAP_NORM(regret[row, column]) + text_color = "white" if normalized > 0.66 else INK + ax.text( + column, + row, + f"{value:.3f}", + ha="center", + va="center", + fontsize=7.7, + color=text_color, + fontweight="bold" if row == column else "normal", + ) + if row == column: + ax.add_patch( + Rectangle( + (column - 0.48, row - 0.48), + 0.96, + 0.96, + facecolor="none", + edgecolor=accent, + linewidth=1.7, + ) + ) + + ax.set_xticks(np.arange(-0.5, 5, 1), minor=True) + ax.set_yticks(np.arange(-0.5, 5, 1), minor=True) + ax.grid(which="minor", color="white", linewidth=1.25) + ax.tick_params(which="minor", bottom=False, left=False) + for spine in ax.spines.values(): + spine.set_visible(False) + return image + + +def draw_gain_table( + ax: plt.Axes, + matrix: np.ndarray, + generalist: np.ndarray, + accent: str, +) -> None: + diagonal = np.diag(matrix) + delta = diagonal - generalist + percent = 100.0 * delta / generalist + if not np.all(delta > 0): + raise ValueError("matched-domain MoS does not improve every domain") + + mean_generalist = float(np.mean(generalist)) + mean_diagonal = float(np.mean(diagonal)) + mean_delta = mean_diagonal - mean_generalist + mean_percent = 100.0 * mean_delta / mean_generalist + + labels = DOMAIN_LABELS + ["Mean"] + deltas = np.concatenate([delta, [mean_delta]]) + percents = np.concatenate([percent, [mean_percent]]) + + ax.set_xlim(0.0, 1.0) + ax.set_ylim(0.0, 1.0) + ax.axis("off") + ax.text( + 0.02, + 0.965, + "Matched MLP gain vs Generalist", + ha="left", + va="top", + fontsize=9.1, + fontweight="bold", + color=INK, + ) + ax.text(0.02, 0.855, "Domain", ha="left", va="center", color=MUTED, fontweight="bold") + ax.text(0.68, 0.855, "Δ AL", ha="right", va="center", color=MUTED, fontweight="bold") + ax.text(0.98, 0.855, "Δ %", ha="right", va="center", color=MUTED, fontweight="bold") + ax.plot([0.02, 0.98], [0.815, 0.815], color=RULE, lw=0.9) + + ys = np.linspace(0.735, 0.175, len(labels)) + for index, (label, value, pct, y) in enumerate(zip(labels, deltas, percents, ys)): + if index == len(labels) - 1: + ax.add_patch( + Rectangle( + (0.01, y - 0.050), + 0.98, + 0.100, + facecolor=ROW_FILL, + edgecolor="none", + zorder=0, + ) + ) + weight = "bold" if index == len(labels) - 1 else "normal" + ax.text(0.02, y, label, ha="left", va="center", color=INK, fontweight=weight) + ax.text( + 0.68, + y, + f"+{value:.3f}", + ha="right", + va="center", + color=accent, + fontweight="bold", + ) + ax.text( + 0.98, + y, + f"+{pct:.1f}%", + ha="right", + va="center", + color=accent, + fontweight="bold", + ) + + ax.text( + 0.02, + 0.045, + "Mean is unweighted across the five domains.", + ha="left", + va="bottom", + fontsize=6.7, + color=MUTED, + ) + + +def render_one( + matrix: np.ndarray, + generalist: np.ndarray, + title: str, + subtitle: str, + cmap: LinearSegmentedColormap, + accent: str, + output: Path, +) -> None: + fig = plt.figure(figsize=(7.15, 3.55), facecolor="white") + grid = fig.add_gridspec( + 1, + 2, + width_ratios=[1.20, 0.92], + wspace=0.22, + left=0.085, + right=0.985, + top=0.755, + bottom=0.21, + ) + ax_matrix = fig.add_subplot(grid[0, 0]) + ax_gain = fig.add_subplot(grid[0, 1]) + + image = draw_matrix(ax_matrix, matrix, cmap, accent) + draw_gain_table(ax_gain, matrix, generalist, accent) + + fig.text(0.03, 0.970, title, ha="left", va="top", fontsize=11.2, fontweight="bold", color=INK) + fig.text(0.03, 0.862, subtitle, ha="left", va="top", fontsize=7.2, color=MUTED) + + cbar_ax = fig.add_axes([0.137, 0.095, 0.355, 0.018]) + cbar = fig.colorbar(image, cax=cbar_ax, orientation="horizontal") + cbar.set_ticks([-1.2, -0.6, 0.0], labels=["−1.2", "−0.6", "0"]) + cbar.ax.tick_params(labelsize=6.5, length=2.0, color=RULE, pad=1.5) + cbar.outline.set_visible(False) + fig.text( + 0.314, + 0.040, + "Cell shade: AL difference from the matched MLP in each column", + ha="center", + va="bottom", + fontsize=6.5, + color=MUTED, + ) + + output.parent.mkdir(parents=True, exist_ok=True) + fig.savefig(output, dpi=420, facecolor="white") + plt.close(fig) + print(f"saved {output}") + + +def render_generalist( + generalist: np.ndarray, + subtitle: str, + output: Path, +) -> None: + mean_al = float(np.mean(generalist)) + x = np.arange(len(DOMAINS)) + + fig, ax = plt.subplots(figsize=(7.15, 3.35), facecolor="white") + fig.subplots_adjust(left=0.095, right=0.975, top=0.755, bottom=0.205) + bars = ax.bar( + x, + generalist, + width=0.58, + color=GENERALIST, + edgecolor=INK, + linewidth=0.55, + zorder=3, + ) + ax.bar_label( + bars, + labels=[f"{value:.3f}" for value in generalist], + padding=-16, + fontsize=8.1, + fontweight="bold", + color="white", + ) + ax.axhline( + mean_al, + color=INK, + lw=1.15, + ls=(0, (4, 2)), + label=f"Five-domain mean = {mean_al:.3f}", + zorder=2, + ) + + ax.set_xlim(-0.55, len(DOMAINS) - 0.45) + ax.set_ylim(0.0, 5.55) + ax.set_xticks(x, labels=DOMAIN_LABELS) + ax.set_yticks(np.arange(0.0, 5.6, 1.0)) + ax.set_ylabel("Acceptance length (AL)", fontweight="bold") + ax.grid(axis="y", color=RULE, linewidth=0.65, zorder=0) + ax.legend(loc="upper right", frameon=False, fontsize=7.4, handlelength=2.8) + ax.spines["top"].set_visible(False) + ax.spines["right"].set_visible(False) + ax.spines["left"].set_color(RULE) + ax.spines["bottom"].set_color(RULE) + ax.tick_params(color=RULE, labelcolor=INK, width=0.65, length=2.8) + + fig.text( + 0.03, + 0.970, + "Generalist (DFlash baseline): AL across five domains", + ha="left", + va="top", + fontsize=11.2, + fontweight="bold", + color=INK, + ) + fig.text(0.03, 0.862, subtitle, ha="left", va="top", fontsize=7.2, color=MUTED) + + output.parent.mkdir(parents=True, exist_ok=True) + fig.savefig(output, dpi=420, facecolor="white") + plt.close(fig) + print(f"saved {output}") + + +def draw_compact_generalist(ax: plt.Axes, generalist: np.ndarray) -> None: + x = np.arange(len(DOMAINS)) + bars = ax.bar( + x, + generalist, + width=0.66, + color=GENERALIST, + edgecolor=INK, + linewidth=0.45, + zorder=3, + ) + ax.bar_label( + bars, + labels=[f"{value:.3f}" for value in generalist], + padding=-10, + fontsize=5.7, + fontweight="bold", + color="white", + ) + ax.axhline(float(np.mean(generalist)), color=INK, lw=0.85, ls=(0, (3, 2)), zorder=2) + ax.set_xlim(-0.55, len(DOMAINS) - 0.45) + ax.set_ylim(0.0, 5.55) + ax.set_xticks(x, labels=["Code", "Math", "FQA", "Creat.", "Gen."]) + ax.tick_params(axis="x", rotation=40, labelsize=5.5, pad=1.8) + ax.set_yticks([0, 2, 4], labels=["0", "2", "4"]) + ax.tick_params(axis="y", labelsize=5.5) + ax.set_ylabel("AL", fontsize=6.5, fontweight="bold", labelpad=2.0) + ax.grid(axis="y", color=RULE, linewidth=0.5, zorder=0) + ax.spines["top"].set_visible(False) + ax.spines["right"].set_visible(False) + ax.spines["left"].set_color(RULE) + ax.spines["bottom"].set_color(RULE) + ax.tick_params(color=RULE, labelcolor=INK, width=0.5, length=2.0) + ax.text( + 0.98, + 0.96, + f"mean {np.mean(generalist):.3f}", + transform=ax.transAxes, + ha="right", + va="top", + fontsize=5.8, + color=INK, + fontweight="bold", + ) + + +def draw_compact_matrix( + ax: plt.Axes, + matrix: np.ndarray, + cmap: LinearSegmentedColormap, + accent: str, +) -> None: + regret = matrix - np.diag(matrix)[None, :] + ax.imshow(regret, cmap=cmap, norm=HEATMAP_NORM, aspect="equal") + short_labels = ["Code", "Math", "FQA", "Creat.", "Gen."] + ax.set_xticks(range(5), labels=short_labels) + ax.set_yticks(range(5), labels=short_labels) + ax.tick_params(axis="x", rotation=40, length=0, pad=1.8, labelsize=5.3) + ax.tick_params(axis="y", length=0, pad=2.0, labelsize=5.3) + ax.set_ylabel("Selected MLP", fontsize=6.1, fontweight="bold", labelpad=2.0) + + for row in range(5): + for column in range(5): + normalized = HEATMAP_NORM(regret[row, column]) + ax.text( + column, + row, + f"{matrix[row, column]:.2f}", + ha="center", + va="center", + fontsize=5.4, + color="white" if normalized > 0.66 else INK, + fontweight="bold" if row == column else "normal", + ) + if row == column: + ax.add_patch( + Rectangle( + (column - 0.47, row - 0.47), + 0.94, + 0.94, + facecolor="none", + edgecolor=accent, + linewidth=1.15, + ) + ) + + ax.set_xticks(np.arange(-0.5, 5, 1), minor=True) + ax.set_yticks(np.arange(-0.5, 5, 1), minor=True) + ax.grid(which="minor", color="white", linewidth=0.9) + ax.tick_params(which="minor", bottom=False, left=False) + for spine in ax.spines.values(): + spine.set_visible(False) + + +def draw_compact_gains( + ax: plt.Axes, + matrix: np.ndarray, + generalist: np.ndarray, + accent: str, +) -> None: + delta = np.diag(matrix) - generalist + percent = 100.0 * delta / generalist + mean_delta = float(np.mean(np.diag(matrix)) - np.mean(generalist)) + mean_percent = 100.0 * mean_delta / float(np.mean(generalist)) + + ax.set_xlim(0.0, 1.0) + ax.set_ylim(4.5, -0.5) + ax.axis("off") + ax.text(0.43, 1.045, "ΔAL", transform=ax.transAxes, ha="right", va="bottom", fontsize=5.5, color=MUTED, fontweight="bold") + ax.text(0.98, 1.045, "Δ%", transform=ax.transAxes, ha="right", va="bottom", fontsize=5.5, color=MUTED, fontweight="bold") + for row, (value, pct) in enumerate(zip(delta, percent)): + ax.text(0.43, row, f"+{value:.2f}", ha="right", va="center", fontsize=5.4, color=accent, fontweight="bold") + ax.text(0.98, row, f"+{pct:.1f}", ha="right", va="center", fontsize=5.4, color=accent, fontweight="bold") + ax.text( + 0.98, + -0.16, + f"mean +{mean_delta:.2f} / +{mean_percent:.1f}%", + transform=ax.transAxes, + ha="right", + va="top", + fontsize=5.1, + color=accent, + fontweight="bold", + ) + + +def render_three_panel( + generalist: np.ndarray, + d0_matrix: np.ndarray, + warm_matrix: np.ndarray, + output: Path, +) -> None: + fig = plt.figure(figsize=(7.15, 2.48), facecolor="white") + outer = fig.add_gridspec( + 1, + 3, + width_ratios=[0.78, 1.36, 1.36], + wspace=0.30, + left=0.055, + right=0.992, + top=0.78, + bottom=0.23, + ) + ax_a = fig.add_subplot(outer[0, 0]) + grid_b = outer[0, 1].subgridspec(1, 2, width_ratios=[1.0, 0.42], wspace=0.04) + ax_b = fig.add_subplot(grid_b[0, 0]) + ax_b_gain = fig.add_subplot(grid_b[0, 1]) + grid_c = outer[0, 2].subgridspec(1, 2, width_ratios=[1.0, 0.42], wspace=0.04) + ax_c = fig.add_subplot(grid_c[0, 0]) + ax_c_gain = fig.add_subplot(grid_c[0, 1]) + + draw_compact_generalist(ax_a, generalist) + draw_compact_matrix(ax_b, d0_matrix, D0_CMAP, D0) + draw_compact_gains(ax_b_gain, d0_matrix, generalist, D0) + draw_compact_matrix(ax_c, warm_matrix, WARM_CMAP, WARM) + draw_compact_gains(ax_c_gain, warm_matrix, generalist, WARM) + + panel_titles = ( + (0.055, "A", "Generalist (DFlash)"), + (0.305, "B", "DFlash-init MoS"), + (0.661, "C", "Generalist-warm-start MoS"), + ) + for x, letter, title in panel_titles: + fig.text(x, 0.935, letter, ha="left", va="top", fontsize=8.8, fontweight="bold", color=INK) + fig.text(x + 0.025, 0.935, title, ha="left", va="top", fontsize=8.0, fontweight="bold", color=INK) + + fig.text( + 0.63, + 0.055, + "Rows select MLPs; columns are evaluation domains. Bold diagonal = matched MLP; gains are vs Generalist.", + ha="center", + va="bottom", + fontsize=5.3, + color=MUTED, + ) + fig.text( + 0.055, + 0.055, + "Qwen3-8B target · fixed seed", + ha="left", + va="bottom", + fontsize=5.3, + color=MUTED, + ) + + output.parent.mkdir(parents=True, exist_ok=True) + fig.savefig(output, dpi=480, facecolor="white") + plt.close(fig) + print(f"saved {output}") + + +def draw_baseline_aligned_panel( + ax_matrix: plt.Axes, + ax_gain: plt.Axes, + matrix: np.ndarray, + generalist: np.ndarray, +) -> None: + aligned = np.vstack([generalist, matrix]) + row_labels = ["Generalist", "Code MLP", "Math MLP", "FQA MLP", "Creat. MLP", "Gen. MLP"] + column_labels = ["Code", "Math", "FQA", "Creat.", "Gen."] + ax_matrix.imshow(aligned, cmap=ABSOLUTE_CMAP, norm=ABSOLUTE_NORM, aspect="equal") + ax_matrix.set_xticks(range(5), labels=column_labels) + ax_matrix.set_yticks(range(6), labels=row_labels) + ax_matrix.tick_params(axis="x", rotation=37, length=0, pad=2.0, labelsize=5.5) + ax_matrix.tick_params(axis="y", length=0, pad=2.4, labelsize=5.4) + + for row in range(6): + for column in range(5): + value = aligned[row, column] + normalized = ABSOLUTE_NORM(value) + is_matched = row > 0 and row - 1 == column + ax_matrix.text( + column, + row, + f"{value:.2f}", + ha="center", + va="center", + fontsize=5.7, + color="white" if normalized > 0.58 else INK, + fontweight="bold" if is_matched else "normal", + ) + if is_matched: + ax_matrix.add_patch( + Rectangle( + (column - 0.47, row - 0.47), + 0.94, + 0.94, + facecolor="none", + edgecolor=INK, + linewidth=1.0, + ) + ) + + ax_matrix.axhline(0.5, color=INK, lw=1.15) + ax_matrix.set_xticks(np.arange(-0.5, 5, 1), minor=True) + ax_matrix.set_yticks(np.arange(-0.5, 6, 1), minor=True) + ax_matrix.grid(which="minor", color="white", linewidth=0.9) + ax_matrix.tick_params(which="minor", bottom=False, left=False) + for spine in ax_matrix.spines.values(): + spine.set_visible(False) + + delta = np.diag(matrix) - generalist + percent = 100.0 * delta / generalist + mean_delta = float(np.mean(np.diag(matrix)) - np.mean(generalist)) + mean_percent = 100.0 * mean_delta / float(np.mean(generalist)) + ax_gain.set_xlim(0.0, 1.0) + ax_gain.set_ylim(5.5, -0.5) + ax_gain.axis("off") + ax_gain.text(0.43, 1.04, "ΔAL", transform=ax_gain.transAxes, ha="right", va="bottom", fontsize=5.7, color=MUTED, fontweight="bold") + ax_gain.text(0.98, 1.04, "Δ%", transform=ax_gain.transAxes, ha="right", va="bottom", fontsize=5.7, color=MUTED, fontweight="bold") + ax_gain.text(0.43, 0, "—", ha="right", va="center", fontsize=5.4, color=MUTED) + ax_gain.text(0.98, 0, "—", ha="right", va="center", fontsize=5.4, color=MUTED) + for row, (value, pct) in enumerate(zip(delta, percent), start=1): + ax_gain.text(0.43, row, f"+{value:.2f}", ha="right", va="center", fontsize=5.5, color=INK, fontweight="bold") + ax_gain.text(0.98, row, f"+{pct:.1f}", ha="right", va="center", fontsize=5.5, color=INK, fontweight="bold") + ax_gain.axhline(0.5, color=INK, lw=1.15) + ax_gain.text( + 0.98, + -0.14, + f"mean +{mean_delta:.2f} / +{mean_percent:.1f}%", + transform=ax_gain.transAxes, + ha="right", + va="top", + fontsize=5.2, + color=INK, + fontweight="bold", + ) + + +def render_baseline_aligned( + generalist: np.ndarray, + d0_matrix: np.ndarray, + warm_matrix: np.ndarray, + output: Path, +) -> None: + fig = plt.figure(figsize=(7.15, 2.85), facecolor="white") + outer = fig.add_gridspec( + 1, + 2, + wspace=0.28, + left=0.105, + right=0.992, + top=0.72, + bottom=0.23, + ) + grid_a = outer[0, 0].subgridspec(1, 2, width_ratios=[1.0, 0.35], wspace=0.04) + ax_a = fig.add_subplot(grid_a[0, 0]) + ax_a_gain = fig.add_subplot(grid_a[0, 1]) + grid_b = outer[0, 1].subgridspec(1, 2, width_ratios=[1.0, 0.35], wspace=0.04) + ax_b = fig.add_subplot(grid_b[0, 0]) + ax_b_gain = fig.add_subplot(grid_b[0, 1]) + + draw_baseline_aligned_panel(ax_a, ax_a_gain, d0_matrix, generalist) + draw_baseline_aligned_panel(ax_b, ax_b_gain, warm_matrix, generalist) + + fig.text( + 0.055, + 0.970, + "Generalist-aligned acceptance-length matrices · Qwen3-8B target", + ha="left", + va="top", + fontsize=9.2, + fontweight="bold", + color=INK, + ) + fig.text( + 0.055, + 0.895, + f"Shared DFlash Generalist baseline mean = {np.mean(generalist):.3f}; fixed-seed selected-checkpoint evaluation", + ha="left", + va="top", + fontsize=6.0, + color=MUTED, + ) + fig.text(0.105, 0.805, "A DFlash-init MoS", ha="left", va="top", fontsize=7.5, fontweight="bold", color=INK) + fig.text(0.563, 0.805, "B Generalist-warm-start MoS", ha="left", va="top", fontsize=7.5, fontweight="bold", color=INK) + fig.text( + 0.50, + 0.045, + "The shared Generalist row is repeated for direct comparison; it is one baseline, not five specialists. Bold boxes mark matched MLPs.", + ha="center", + va="bottom", + fontsize=5.2, + color=MUTED, + ) + + output.parent.mkdir(parents=True, exist_ok=True) + fig.savefig(output, dpi=480, facecolor="white") + plt.close(fig) + print(f"saved {output}") + + +def draw_summary_matrix( + ax: plt.Axes, + matrix: np.ndarray, + cmap: LinearSegmentedColormap, + accent: str, +) -> None: + regret = matrix - np.diag(matrix)[None, :] + ax.imshow(regret, cmap=cmap, norm=HEATMAP_NORM, aspect="equal") + labels = ["Code", "Math", "FQA", "Creat.", "Gen."] + ax.set_xticks(range(5), labels=labels) + ax.set_yticks(range(5), labels=labels) + ax.tick_params(axis="x", rotation=38, length=0, pad=2.0, labelsize=5.4) + ax.tick_params(axis="y", length=0, pad=2.2, labelsize=5.4) + ax.set_ylabel("Selected MLP", fontsize=6.1, fontweight="bold", labelpad=2.2) + + for row in range(5): + for column in range(5): + normalized = HEATMAP_NORM(regret[row, column]) + is_matched = row == column + ax.text( + column, + row, + f"{matrix[row, column]:.3f}", + ha="center", + va="center", + fontsize=5.2, + color="white" if normalized > 0.66 else INK, + fontweight="bold" if is_matched else "normal", + ) + if is_matched: + ax.add_patch( + Rectangle( + (column - 0.47, row - 0.47), + 0.94, + 0.94, + facecolor="none", + edgecolor=accent, + linewidth=1.15, + ) + ) + + ax.set_xticks(np.arange(-0.5, 5, 1), minor=True) + ax.set_yticks(np.arange(-0.5, 5, 1), minor=True) + ax.grid(which="minor", color="white", linewidth=0.95) + ax.tick_params(which="minor", bottom=False, left=False) + for spine in ax.spines.values(): + spine.set_visible(False) + + +def draw_three_method_table( + ax: plt.Axes, + generalist: np.ndarray, + d0_matrix: np.ndarray, + warm_matrix: np.ndarray, +) -> None: + d0 = np.diag(d0_matrix) + warm = np.diag(warm_matrix) + labels = DOMAIN_LABELS + ["Mean"] + generalist_values = np.concatenate([generalist, [np.mean(generalist)]]) + d0_values = np.concatenate([d0, [np.mean(d0)]]) + warm_values = np.concatenate([warm, [np.mean(warm)]]) + + ax.set_xlim(0.0, 1.0) + ax.set_ylim(0.0, 1.0) + ax.axis("off") + header_y = 0.875 + ax.text(0.01, header_y, "Domain", ha="left", va="center", fontsize=5.9, color=MUTED, fontweight="bold") + ax.text(0.48, header_y, "Generalist", ha="right", va="center", fontsize=5.7, color=MUTED, fontweight="bold") + ax.text(0.75, header_y, "D0 MoS", ha="right", va="center", fontsize=5.7, color=D0, fontweight="bold") + ax.text(0.99, header_y, "G-init", ha="right", va="center", fontsize=5.7, color=WARM, fontweight="bold") + ax.plot([0.01, 0.99], [0.825, 0.825], color=RULE, lw=0.8) + + ys = np.linspace(0.745, 0.245, len(labels)) + for index, (label, gen_value, d0_value, warm_value, y) in enumerate( + zip(labels, generalist_values, d0_values, warm_values, ys) + ): + is_mean = index == len(labels) - 1 + if is_mean: + ax.add_patch( + Rectangle( + (0.0, y - 0.045), + 1.0, + 0.090, + facecolor=ROW_FILL, + edgecolor="none", + zorder=0, + ) + ) + weight = "bold" if is_mean else "normal" + ax.text(0.01, y, label, ha="left", va="center", fontsize=5.8, color=INK, fontweight=weight) + ax.text(0.48, y, f"{gen_value:.3f}", ha="right", va="center", fontsize=5.8, color=MUTED, fontweight=weight) + ax.text(0.75, y, f"{d0_value:.3f}", ha="right", va="center", fontsize=5.8, color=D0, fontweight="bold") + ax.text(0.99, y, f"{warm_value:.3f}", ha="right", va="center", fontsize=5.8, color=WARM, fontweight="bold") + + d0_delta = float(np.mean(d0) - np.mean(generalist)) + warm_delta = float(np.mean(warm) - np.mean(generalist)) + d0_percent = 100.0 * d0_delta / float(np.mean(generalist)) + warm_percent = 100.0 * warm_delta / float(np.mean(generalist)) + ax.text( + 0.99, + 0.105, + f"D0 mean gain +{d0_delta:.3f} / +{d0_percent:.1f}%", + ha="right", + va="center", + fontsize=5.4, + color=D0, + fontweight="bold", + ) + ax.text( + 0.99, + 0.035, + f"G-init mean gain +{warm_delta:.3f} / +{warm_percent:.1f}%", + ha="right", + va="center", + fontsize=5.4, + color=WARM, + fontweight="bold", + ) + + +def render_matrices_summary( + generalist: np.ndarray, + d0_matrix: np.ndarray, + warm_matrix: np.ndarray, + output: Path, +) -> None: + fig = plt.figure(figsize=(7.15, 2.52), facecolor="white") + grid = fig.add_gridspec( + 1, + 3, + width_ratios=[1.0, 1.0, 1.18], + wspace=0.28, + left=0.065, + right=0.992, + top=0.77, + bottom=0.22, + ) + ax_d0 = fig.add_subplot(grid[0, 0]) + ax_warm = fig.add_subplot(grid[0, 1]) + ax_table = fig.add_subplot(grid[0, 2]) + + draw_summary_matrix(ax_d0, d0_matrix, D0_CMAP, D0) + draw_summary_matrix(ax_warm, warm_matrix, WARM_CMAP, WARM) + draw_three_method_table(ax_table, generalist, d0_matrix, warm_matrix) + + titles = ( + (0.065, "A", "DFlash-init MoS"), + (0.360, "B", "Generalist-warm-start MoS"), + (0.670, "C", "Matched-domain AL"), + ) + for x, letter, title in titles: + fig.text(x, 0.940, letter, ha="left", va="top", fontsize=8.7, fontweight="bold", color=INK) + fig.text(x + 0.025, 0.940, title, ha="left", va="top", fontsize=7.5, fontweight="bold", color=INK) + + fig.text( + 0.50, + 0.045, + "Qwen3-8B target · fixed-seed selected checkpoints · matrix columns are evaluation domains; bold diagonal cells select the matched MLP.", + ha="center", + va="bottom", + fontsize=5.2, + color=MUTED, + ) + + output.parent.mkdir(parents=True, exist_ok=True) + fig.savefig(output, dpi=480, facecolor="white") + plt.close(fig) + print(f"saved {output}") + + +def main() -> None: + args = parse_args() + configure_style() + evidence, generalist = load_evidence(args.evidence) + + d0_matrix = matrix_from_evidence(evidence, "panel_b_matrix_dflash_init") + warm_matrix = matrix_from_evidence(evidence, "panel_c_matrix_warm_start") + + subtitle = ( + "Qwen3-8B target · selected-checkpoint, fixed-seed evaluation · " + "rows: selected MLP; columns: evaluation domain" + ) + render_one( + d0_matrix, + generalist, + "DFlash-initialized MoS: 5×5 routing matrix", + subtitle, + D0_CMAP, + D0, + args.output_dir / "fig_mos_d0_matrix_gains.png", + ) + render_one( + warm_matrix, + generalist, + "Generalist-warm-started MoS: 5×5 routing matrix", + subtitle, + WARM_CMAP, + WARM, + args.output_dir / "fig_mos_ginit_matrix_gains.png", + ) + render_generalist( + generalist, + "Qwen3-8B target · selected-checkpoint, fixed-seed evaluation · standard single-model DFlash", + args.output_dir / "fig_generalist_domain_al.png", + ) + render_three_panel( + generalist, + d0_matrix, + warm_matrix, + args.output_dir / "fig_mos_three_panel.png", + ) + render_baseline_aligned( + generalist, + d0_matrix, + warm_matrix, + args.output_dir / "fig_mos_baseline_aligned.png", + ) + render_matrices_summary( + generalist, + d0_matrix, + warm_matrix, + args.output_dir / "fig_mos_matrices_summary.png", + ) + + +if __name__ == "__main__": + main() diff --git a/recipes/plotting/v1/plot_qwen3_4b_matched.py b/recipes/plotting/v1/plot_qwen3_4b_matched.py new file mode 100644 index 0000000000000000000000000000000000000000..e50d3cd7ff28555d3eb9a98526398e7e25c40900 --- /dev/null +++ b/recipes/plotting/v1/plot_qwen3_4b_matched.py @@ -0,0 +1,199 @@ +#!/usr/bin/env python3 +"""Render the Qwen3-4B matched-volume trajectory for the AAAI supplement.""" + +from __future__ import annotations + +import argparse +import csv +from pathlib import Path + +import matplotlib as mpl +import matplotlib.pyplot as plt +import numpy as np + + +GENERALIST_COLOR = "#5B5B5B" +MOS_COLOR = "#6F5AA8" +GRID_COLOR = "#D9D9D9" + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument( + "--input", + type=Path, + default=Path( + "paper/submission/evidence/b5_qwen3_4b/" + "matched_volume_trajectory.csv" + ), + ) + parser.add_argument( + "--output", + type=Path, + default=Path("paper/submission/figures/fig_qwen3_4b_matched"), + help="Output stem; both PDF and PNG are written.", + ) + return parser.parse_args() + + +def load_rows(path: Path) -> dict[str, np.ndarray]: + with path.open(newline="") as handle: + rows = list(csv.DictReader(handle)) + if len(rows) != 29: + raise ValueError(f"expected 29 matched points, found {len(rows)}") + + keys = ( + "training_samples", + "generalist_overall_al", + "arm_a_overall_al", + "delta_overall_al", + ) + arrays = { + key: np.asarray([float(row[key]) for row in rows], dtype=np.float64) + for key in keys + } + if not np.all(np.diff(arrays["training_samples"]) > 0): + raise ValueError("training_samples must be strictly increasing") + recomputed = arrays["arm_a_overall_al"] - arrays["generalist_overall_al"] + if not np.allclose(recomputed, arrays["delta_overall_al"], atol=5e-5): + raise ValueError("stored deltas disagree with trajectory values") + if not np.all(arrays["delta_overall_al"] > 0): + raise ValueError("the publication annotation assumes 29/29 positive deltas") + return arrays + + +def configure_style() -> None: + mpl.rcParams.update( + { + "font.family": "serif", + "font.serif": ["Times New Roman", "Times", "Nimbus Roman", "DejaVu Serif"], + "font.size": 8.0, + "axes.labelsize": 8.0, + "axes.titlesize": 8.0, + "xtick.labelsize": 7.2, + "ytick.labelsize": 7.2, + "legend.fontsize": 7.1, + "axes.linewidth": 0.7, + "lines.linewidth": 1.5, + "lines.markersize": 3.4, + "pdf.fonttype": 42, + "ps.fonttype": 42, + "savefig.bbox": "tight", + "savefig.pad_inches": 0.02, + } + ) + + +def render(data: dict[str, np.ndarray], output: Path) -> None: + configure_style() + samples_m = data["training_samples"] / 1_000_000.0 + generalist = data["generalist_overall_al"] + mos = data["arm_a_overall_al"] + delta = data["delta_overall_al"] + median_delta = float(np.median(delta)) + + fig, (ax_curve, ax_delta) = plt.subplots( + 1, + 2, + figsize=(7.0, 2.42), + gridspec_kw={"width_ratios": [1.16, 0.84], "wspace": 0.31}, + ) + + ax_curve.plot( + samples_m, + generalist, + color=GENERALIST_COLOR, + linestyle="--", + marker="o", + markerfacecolor="white", + markeredgewidth=0.75, + markevery=2, + label="Generalist", + zorder=2, + ) + ax_curve.plot( + samples_m, + mos, + color=MOS_COLOR, + linestyle="-", + marker="s", + markerfacecolor=MOS_COLOR, + markeredgewidth=0.0, + markevery=2, + label="D0-MoS (5 groups)", + zorder=3, + ) + ax_curve.set_xlabel("Training samples (millions)") + ax_curve.set_ylabel("Five-domain mean AL") + ax_curve.set_xlim(0.0, 2.4) + ymin = min(float(generalist.min()), float(mos.min())) - 0.025 + ymax = max(float(generalist.max()), float(mos.max())) + 0.025 + ax_curve.set_ylim(ymin, ymax) + ax_curve.grid(axis="y", color=GRID_COLOR, linewidth=0.55, alpha=0.8) + ax_curve.legend(loc="lower right", frameon=False, handlelength=2.2) + + ax_delta.axhline(0.0, color=GENERALIST_COLOR, linewidth=0.75, linestyle=":") + ax_delta.plot( + samples_m, + delta, + color=MOS_COLOR, + linestyle="-", + marker="D", + markerfacecolor="white", + markeredgewidth=0.75, + markevery=2, + zorder=3, + ) + ax_delta.axhline( + median_delta, + color=MOS_COLOR, + linewidth=0.9, + linestyle="--", + alpha=0.8, + ) + ax_delta.text( + 0.98, + 0.08, + f"29/29 matched points > 0\nmedian $\\Delta$ = {median_delta:.3f}", + transform=ax_delta.transAxes, + ha="right", + va="bottom", + fontsize=7.0, + ) + ax_delta.set_xlabel("Training samples (millions)") + ax_delta.set_ylabel(r"$\Delta$ AL (MoS $-$ generalist)") + ax_delta.set_xlim(0.0, 2.4) + ax_delta.set_ylim(0.0, max(0.12, float(delta.max()) + 0.01)) + ax_delta.grid(axis="y", color=GRID_COLOR, linewidth=0.55, alpha=0.8) + + for label, axis in (("(a)", ax_curve), ("(b)", ax_delta)): + axis.text( + -0.14, + 1.03, + label, + transform=axis.transAxes, + ha="left", + va="bottom", + fontweight="bold", + ) + axis.spines["top"].set_visible(False) + axis.spines["right"].set_visible(False) + axis.tick_params(width=0.7, length=3.0) + + output.parent.mkdir(parents=True, exist_ok=True) + metadata = { + "Title": "Qwen3-4B matched-volume MoS replication", + "Subject": "Five-domain acceptance length over matched training volume", + } + fig.savefig(output.with_suffix(".pdf"), metadata=metadata) + fig.savefig(output.with_suffix(".png"), dpi=450) + plt.close(fig) + + +def main() -> None: + args = parse_args() + render(load_rows(args.input), args.output) + + +if __name__ == "__main__": + main() diff --git a/recipes/plotting/v1/plot_recipe_budget.py b/recipes/plotting/v1/plot_recipe_budget.py new file mode 100644 index 0000000000000000000000000000000000000000..57d76f806dc2145362ae969783bc59c2484cc6c7 --- /dev/null +++ b/recipes/plotting/v1/plot_recipe_budget.py @@ -0,0 +1,130 @@ +import matplotlib; matplotlib.use("Agg") +import matplotlib.pyplot as plt +from matplotlib.lines import Line2D +import csv, glob, os +HERE=os.path.dirname(os.path.abspath(__file__)) +E=os.environ.get("MOS_EVAL_DIR",os.path.normpath(os.path.join(HERE,"..","eval_csv"))) +# ---- 颜色宪法(固定,勿改):gen灰 merge橙 route蓝 A绿 B紫 C金 D青 ---- +CG="#8a8f98"; CM="#E8590C"; CR="#1971C2"; CA="#2F9E44"; CB="#9C36B5"; CC="#E8B117"; CD="#0B7285" +CRW="#4DABF7"; CAW="#D6336C"; CSG="#C92A2A"; CSD="#5F3DC4" +def loadal(p): + d={} + for r in csv.reader(open(p)): + if not r or r[0].startswith("ckpt"): continue + try: d[int(r[1])]=float(r[3]) + except: pass + return d +v2=loadal(f"{E}/mv2c89/al_curve.csv"); v3={**loadal(f"{E}/mv3x89/al_curve.csv"),**loadal(f"{E}/merge89/al_curve.csv")} +rc=loadal(f"{E}/rc89/al_curve.csv"); rw={**loadal(f"{E}/rw89x/al_curve.csv"),**loadal(f"{E}/route89/al_curve.csv")} +gen={} +for f in glob.glob(f"{E}/gendense_l*/*.csv"): + for r in csv.reader(open(f)): + if len(r)>=4 and r[2]=="code": + try: gen[int(r[1])]=float(r[3]) + except: pass +S=64/1e6; PRE=3.2; V2E=49936; RB=31210 +gx=[k*S for k in sorted(gen)]; gy=[gen[k] for k in sorted(gen)] +mx=[PRE+k*S for k in sorted(v2)]+[PRE+(V2E+k)*S for k in sorted(v3)]; my=[v2[k] for k in sorted(v2)]+[v3[k] for k in sorted(v3)] +rck=[k for k in sorted(rc) if k<=30000] +rx=[PRE+k*S for k in rck]+[PRE+(RB+k)*S for k in sorted(rw)]; ry=[rc[k] for k in rck]+[rw[k] for k in sorted(rw)] +rx+=[7.19,7.59,7.99]; ry+=[3.487,3.517,3.537] # WR-2 +rx+=[8.39,8.79,9.19]; ry+=[3.518,3.510,3.544] # WR-3 +rx+=[9.59,9.99,10.39]; ry+=[3.518,3.553,3.568] # WR-4 峰 3.568 +rx+=[10.79,11.19,11.59]; ry+=[3.562,3.549,3.542] # WR-5 衰减,确认到顶 +A=[3.312,3.473,3.549,3.610,3.645,3.616,3.655,3.640]; ax_=[0.8*(i+1) for i in range(len(A))] +B=[3.477,3.586,3.626,3.657,3.709,3.719,3.716,3.710]; bx_=[PRE+0.8*(i+1) for i in range(len(B))] +C=[3.185,3.315,3.395,3.420,3.433,3.440,3.461,3.473,3.466,3.471,3.468]; cx_=[0.8*(i+1) for i in range(11)] +D=[3.471,3.536,3.553,3.606,3.629,3.611,3.648,3.626]; dx_=[PRE+0.8*(i+1) for i in range(8)] +RW=[3.466,3.463,3.473,3.541,3.504,3.549,3.557,3.563,3.550,3.563,3.564,3.562,3.556,3.579,3.589,3.574,3.589]; rwx=[3.2+0.4*(i+1) for i in range(17)] +AW=[3.411,3.391,3.438,3.457,3.446,3.472,3.471,3.462]; awx=[3.2+0.4*(i+1) for i in range(8)] +SG=[3.411,3.490,3.519,3.587,3.563,3.607,3.618,3.581]; sgx=[3.2+0.4*(i+1) for i in range(8)] +SD=[3.186,3.295,3.372,3.421,3.448,3.514,3.488,3.487]; sdx=[0.4*(i+1) for i in range(8)] +# AW/SD are retained for reproducibility but omitted from this compact recipe +# overview; the most relevant G-init. complete code specialist is shown directly, +# while the remaining controls stay in the paper table and supplement. + +# 每条曲线:(x, y, 颜色, marker, 线型, 线宽)；颜色仅作辅助，灰度下由 marker/线型区分 +E2E=2.6; CTX=1.7 # 端到端加粗,其余作背景 +CURVES=[ + (gx,gy,CG,"s",":",CTX), + (mx,my,CM,"P","-.",CTX), + (rx,ry,CR,"o","--",CTX), + (rwx,RW,CRW,"X",(0,(6,2)),CTX), + (cx_,C,CC,"D",(0,(3,1,1,1)),E2E), + (dx_,D,CD,"v",(0,(6,2)),E2E), + (sgx,SG,CSG,"^",(0,(1,1)),CTX), + (ax_,A,CA,"o","-",E2E), + (bx_,B,CB,"s","-",E2E), +] + +fig,ax=plt.subplots(figsize=(6.8,3.8)) +for sp in ["top","right"]: ax.spines[sp].set_visible(False) +for sp in ["left","bottom"]: + ax.spines[sp].set_color("#666") + ax.spines[sp].set_linewidth(0.6) +ax.grid(axis="y",alpha=0.55,lw=0.5,color="#c9ced6") +ax.tick_params(colors="#333",labelsize=9.2,width=0.6,length=3) + +# generalist 预训练竖线 + 从 generalist 热启的三条连线(虚线) +ax.axvline(PRE,ls=(0,(2,3)),lw=1.1,color="#c2c7cf") +ax.text(PRE-0.05,3.785,"3.2M pretraining",color="#333",fontsize=9.0,ha="right") +G0=(3.2,3.4075) +for x1,y1,cc in [(bx_[0],B[0],CB),(dx_[0],D[0],CD),(rwx[0],RW[0],CRW)]: + ax.plot([G0[0],x1],[G0[1],y1],ls=(0,(2,2)),lw=1.1,color=cc,alpha=0.45,zorder=2) +ax.plot([G0[0]],[G0[1]],"o",ms=5,color="#8a8f98",mfc="white",mew=1.2,zorder=3) + +for xs,ys,c,mk,ls,lw in CURVES: + ax.plot(xs,ys,color=c,ls=ls,marker=mk,lw=lw,ms=5.0 if lw==E2E else 4.0, + mfc="white",mec="#222",mew=0.8,alpha=1.0,zorder=5 if lw==E2E else 3) + pk=max(ys); i=ys.index(pk) + ax.plot([xs[i]],[pk],marker="*",ls="none",ms=7.5,mfc=c,mec="#222",mew=0.6,zorder=6) + +ax.set_xlim(0,11.9); ax.set_ylim(3.10,3.80) +ax.set_xlabel("Cumulative training samples (millions)",fontsize=10.0) +ax.set_ylabel("Code acceptance length",fontsize=10.0) + +# ---- 右下角:结果排行图例表(按峰值降序) ---- +# (颜色, marker, 线型, 线宽, 名称, 峰值字符串, 是否冠军) +ROWS=[ + (CB,"s","-",E2E,"G-init. MoS, shared updated","3.719",True), + (CA,"o","-",E2E,"D0-init. MoS, shared updated","3.655",False), + (CD,"v",(0,(6,2)),E2E,"G-init. MoS, shared frozen","3.648",False), + (CSG,"^",(0,(1,1)),CTX,"G-init. complete code specialist","3.618",False), + (CRW,"X",(0,(6,2)),CTX,"Frozen generalist, G-init. MLPs","3.589",False), + (CR,"o","--",CTX,"Frozen generalist, D0-init. MLPs","3.568",False), + (CC,"D",(0,(3,1,1,1)),E2E,"D0-init. MoS, shared frozen","3.473",False), + (CM,"P","-.",CTX,"RegMean-style attention merge","3.470",False), + (CG,"s",":",CTX,"Generalist","3.435",False), +] +LX0,LX1=0.455,0.995; LTOP,LBOT=0.455,0.025 +n=len(ROWS); hh=0.060 # header 高 +rh=(LTOP-LBOT-hh)/n +# 背景框 +ax.add_patch(plt.Rectangle((LX0,LBOT),LX1-LX0,LTOP-LBOT,transform=ax.transAxes, + facecolor="#fbfcfd",edgecolor="#dfe3e8",lw=1.0,zorder=8,clip_on=False, + joinstyle="round")) +# 表头 +hy=LTOP-hh*0.58 +ax.text(LX0+0.090,hy,"Recipe",transform=ax.transAxes, + fontsize=9.0,color="#222",fontweight="bold",va="center",zorder=9) +ax.text(LX1-0.018,hy,"Peak AL",transform=ax.transAxes,fontsize=9.0,color="#222", + fontweight="bold",va="center",ha="right",zorder=9) +ax.plot([LX0+0.015,LX1-0.015],[LTOP-hh,LTOP-hh],transform=ax.transAxes, + color="#dfe3e8",lw=1.0,zorder=9,clip_on=False) +# 行 +for j,(c,mk,ls,lw,name,pk,champ) in enumerate(ROWS): + yy=LTOP-hh-rh*(j+0.5) + if champ: + ax.add_patch(plt.Rectangle((LX0+0.006,yy-rh*0.46),LX1-LX0-0.012,rh*0.92, + transform=ax.transAxes,facecolor="#eceff3",edgecolor="none",zorder=8.5,clip_on=False)) + # 色样(短线+marker) + ax.plot([LX0+0.022,LX0+0.068],[yy,yy],color=c,ls=ls,marker=mk,lw=lw,ms=5.0, + mfc="white",mec="#222",mew=0.8,transform=ax.transAxes,zorder=9,clip_on=False) + ax.text(LX0+0.090,yy,name,transform=ax.transAxes,fontsize=9.0,color="#222", + va="center",zorder=9,fontweight="bold" if champ else "normal") + ax.text(LX1-0.018,yy,pk,transform=ax.transAxes,fontsize=9.0,color="#222", + fontweight="bold",va="center",ha="right",zorder=9) + +OUT=os.environ.get("MOS_RECIPE_BUDGET_OUT",os.path.normpath(os.path.join(HERE,"..","figs","recipe_budget.png"))) +plt.tight_layout(pad=0.3); plt.savefig(OUT,dpi=330,bbox_inches="tight") +print("saved") diff --git a/recipes/requirements-v1.txt b/recipes/requirements-v1.txt new file mode 100644 index 0000000000000000000000000000000000000000..59ca10d0a481653d1cb8c08720df801fc81fcea5 --- /dev/null +++ b/recipes/requirements-v1.txt @@ -0,0 +1,4 @@ +matplotlib +numpy +safetensors +torch diff --git a/results/historical-al-curves/v1/a2_code/al_curve.csv b/results/historical-al-curves/v1/a2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..155916b133833863d3ba2f259f06fe0e6f3fd80d --- /dev/null +++ b/results/historical-al-curves/v1/a2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a2_experts89/code,0,code,3.5489096573208725 diff --git a/results/historical-al-curves/v1/a2_creative_writing/al_curve.csv b/results/historical-al-curves/v1/a2_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..4e6195d9e5caaf13d0aa6e5991fe6c5361a3c8b0 --- /dev/null +++ b/results/historical-al-curves/v1/a2_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a2_experts89/creative_writing,0,creative_writing,2.8214548126377665 diff --git a/results/historical-al-curves/v1/a2_factual_qa/al_curve.csv b/results/historical-al-curves/v1/a2_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..1f50e6180ea1c8cd2cf9d7087049981edb15c1e1 --- /dev/null +++ b/results/historical-al-curves/v1/a2_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a2_experts89/factual_qa,0,factual_qa,2.9443757725587143 diff --git a/results/historical-al-curves/v1/a2_general/al_curve.csv b/results/historical-al-curves/v1/a2_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6fb9e1ec6148cbbf291534658fd5ccf79e7531cb --- /dev/null +++ b/results/historical-al-curves/v1/a2_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a2_experts89/general,0,general,3.153636255783604 diff --git a/results/historical-al-curves/v1/a2_math/al_curve.csv b/results/historical-al-curves/v1/a2_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..d14955c1624af575b75e5f5d891229a878a364e1 --- /dev/null +++ b/results/historical-al-curves/v1/a2_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a2_experts89/math,0,math,5.179190751445087 diff --git a/results/historical-al-curves/v1/a3_code/al_curve.csv b/results/historical-al-curves/v1/a3_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..fae5fa5ac71ba8da207bb46eca492b544f78aa6b --- /dev/null +++ b/results/historical-al-curves/v1/a3_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a3_experts89/code,0,code,3.6102044050071305 diff --git a/results/historical-al-curves/v1/a3_creative_writing/al_curve.csv b/results/historical-al-curves/v1/a3_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..f734cc051e644ef0b4d353f3da5305fba07cf85f --- /dev/null +++ b/results/historical-al-curves/v1/a3_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a3_experts89/creative_writing,0,creative_writing,2.8948360346777235 diff --git a/results/historical-al-curves/v1/a3_factual_qa/al_curve.csv b/results/historical-al-curves/v1/a3_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..b9e9c1911a391adc1a27893ed807d7d01d297ebd --- /dev/null +++ b/results/historical-al-curves/v1/a3_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a3_experts89/factual_qa,0,factual_qa,2.9509050334738407 diff --git a/results/historical-al-curves/v1/a3_general/al_curve.csv b/results/historical-al-curves/v1/a3_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..b7c73e66fc897ec62465a232bb232665d0cc3ab9 --- /dev/null +++ b/results/historical-al-curves/v1/a3_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a3_experts89/general,0,general,3.1710199739983453 diff --git a/results/historical-al-curves/v1/a3_math/al_curve.csv b/results/historical-al-curves/v1/a3_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..01a56a8b26d55fa892b3044ceb6f8ac45a04be35 --- /dev/null +++ b/results/historical-al-curves/v1/a3_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a3_experts89/math,0,math,5.330160618679358 diff --git a/results/historical-al-curves/v1/a4_code/al_curve.csv b/results/historical-al-curves/v1/a4_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..3ffb0a71c6ca67f747b7c5bd43609ada844af953 --- /dev/null +++ b/results/historical-al-curves/v1/a4_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a4_experts89/code,0,code,3.6448568229083347 diff --git a/results/historical-al-curves/v1/a4_creative_writing/al_curve.csv b/results/historical-al-curves/v1/a4_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..16cf180db66b01f080b5bd3daced4063258ce46d --- /dev/null +++ b/results/historical-al-curves/v1/a4_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a4_experts89/creative_writing,0,creative_writing,2.9218185276773827 diff --git a/results/historical-al-curves/v1/a4_factual_qa/al_curve.csv b/results/historical-al-curves/v1/a4_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..0e76595ba53852f75f8bd0b7abf5ed48f63f2787 --- /dev/null +++ b/results/historical-al-curves/v1/a4_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a4_experts89/factual_qa,0,factual_qa,2.9440940012368584 diff --git a/results/historical-al-curves/v1/a4_general/al_curve.csv b/results/historical-al-curves/v1/a4_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..0d52cb205674a0aa318b89ef205bfb3f2c137a12 --- /dev/null +++ b/results/historical-al-curves/v1/a4_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a4_experts89/general,0,general,3.2077452044878756 diff --git a/results/historical-al-curves/v1/a4_math/al_curve.csv b/results/historical-al-curves/v1/a4_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..8c10895b8f46fd07d6edab641326473df9a0dd8b --- /dev/null +++ b/results/historical-al-curves/v1/a4_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a4_experts89/math,0,math,5.3365098272781415 diff --git a/results/historical-al-curves/v1/a5_code/al_curve.csv b/results/historical-al-curves/v1/a5_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..395f82e77528c15e2c460c1345d17b33b09e3547 --- /dev/null +++ b/results/historical-al-curves/v1/a5_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a5_experts89/code,0,code,3.6159339787335343 diff --git a/results/historical-al-curves/v1/a5_creative_writing/al_curve.csv b/results/historical-al-curves/v1/a5_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..bdf6b514a764f7c39ce5602927c61f788c87ecc7 --- /dev/null +++ b/results/historical-al-curves/v1/a5_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a5_experts89/creative_writing,0,creative_writing,2.920152091254753 diff --git a/results/historical-al-curves/v1/a5_factual_qa/al_curve.csv b/results/historical-al-curves/v1/a5_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..5e84b183ff178c40acb3c7cbc03e005711025209 --- /dev/null +++ b/results/historical-al-curves/v1/a5_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a5_experts89/factual_qa,0,factual_qa,2.96844206062118 diff --git a/results/historical-al-curves/v1/a5_general/al_curve.csv b/results/historical-al-curves/v1/a5_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..3257be7fdffbdd1e4f46a4c71e5454f22fd18bcc --- /dev/null +++ b/results/historical-al-curves/v1/a5_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a5_experts89/general,0,general,3.2156340755082287 diff --git a/results/historical-al-curves/v1/a5_math/al_curve.csv b/results/historical-al-curves/v1/a5_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..47b7af94226193fc316d5aa0d9f79c71e9f0ccd4 --- /dev/null +++ b/results/historical-al-curves/v1/a5_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a5_experts89/math,0,math,5.334921107472462 diff --git a/results/historical-al-curves/v1/a6_code/al_curve.csv b/results/historical-al-curves/v1/a6_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2a03e78b084f61e815c11da643f787a3c8f7b293 --- /dev/null +++ b/results/historical-al-curves/v1/a6_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a6_experts89/code,0,code,3.655089436111334 diff --git a/results/historical-al-curves/v1/a6_creative_writing/al_curve.csv b/results/historical-al-curves/v1/a6_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..afaa024576ed3850851d945a561afd048b41ff58 --- /dev/null +++ b/results/historical-al-curves/v1/a6_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a6_experts89/creative_writing,0,creative_writing,2.927945101029356 diff --git a/results/historical-al-curves/v1/a6_factual_qa/al_curve.csv b/results/historical-al-curves/v1/a6_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..199539933f753439c01524bf2743ce17e0097a9f --- /dev/null +++ b/results/historical-al-curves/v1/a6_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a6_experts89/factual_qa,0,factual_qa,2.9658779576587797 diff --git a/results/historical-al-curves/v1/a6_general/al_curve.csv b/results/historical-al-curves/v1/a6_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..5155647ab03bfbbfaf81d95b4abb99b1848d376d --- /dev/null +++ b/results/historical-al-curves/v1/a6_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a6_experts89/general,0,general,3.1881461889877376 diff --git a/results/historical-al-curves/v1/a6_math/al_curve.csv b/results/historical-al-curves/v1/a6_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..1a506185242646d0f950ce7f7a202c98bf462464 --- /dev/null +++ b/results/historical-al-curves/v1/a6_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a6_experts89/math,0,math,5.4057315233785825 diff --git a/results/historical-al-curves/v1/a7_code/al_curve.csv b/results/historical-al-curves/v1/a7_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..43bdae9bfdb733c2af8bf109371dd469df518225 --- /dev/null +++ b/results/historical-al-curves/v1/a7_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a7_experts89/code,0,code,3.640198114714811 diff --git a/results/historical-al-curves/v1/a7_creative_writing/al_curve.csv b/results/historical-al-curves/v1/a7_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..09133520ca7d94290fbbe336bf8f5962ebe4d1ec --- /dev/null +++ b/results/historical-al-curves/v1/a7_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a7_experts89/creative_writing,0,creative_writing,2.93297689516899 diff --git a/results/historical-al-curves/v1/a7_factual_qa/al_curve.csv b/results/historical-al-curves/v1/a7_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..141ab69ce0c0e9eeb0d9646d2343421b916842ab --- /dev/null +++ b/results/historical-al-curves/v1/a7_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a7_experts89/factual_qa,0,factual_qa,3.038959374599513 diff --git a/results/historical-al-curves/v1/a7_general/al_curve.csv b/results/historical-al-curves/v1/a7_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..21fdaa6ecd946e15f4c4a913b148cd793dfcdddb --- /dev/null +++ b/results/historical-al-curves/v1/a7_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a7_experts89/general,0,general,3.200690805145307 diff --git a/results/historical-al-curves/v1/a7_math/al_curve.csv b/results/historical-al-curves/v1/a7_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..ce3e98e6521d556c7a7e1d2c53a019ea97514d0b --- /dev/null +++ b/results/historical-al-curves/v1/a7_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +a7_experts89/math,0,math,5.407362703681351 diff --git a/results/historical-al-curves/v1/aw0_code/al_curve.csv b/results/historical-al-curves/v1/aw0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..5a77e6769cafe90c9c6ecb0816a94f822d55d357 --- /dev/null +++ b/results/historical-al-curves/v1/aw0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_attnwarm_code/epoch_0_step_6242,6242,code,3.411033760011977 diff --git a/results/historical-al-curves/v1/aw1_code/al_curve.csv b/results/historical-al-curves/v1/aw1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..99c0a148b1f95da47e959c234e32cac275824a60 --- /dev/null +++ b/results/historical-al-curves/v1/aw1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_attnwarm_code/epoch_1_step_12484,12484,code,3.390980800714392 diff --git a/results/historical-al-curves/v1/aw2_code/al_curve.csv b/results/historical-al-curves/v1/aw2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..ccbaf32664e157ea014d84f9b73d447e1d5af6aa --- /dev/null +++ b/results/historical-al-curves/v1/aw2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_attnwarm_code/epoch_2_step_18726,18726,code,3.4380564357929684 diff --git a/results/historical-al-curves/v1/aw3_code/al_curve.csv b/results/historical-al-curves/v1/aw3_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..7ed7366924f35fbed26eb8319d198177dab84971 --- /dev/null +++ b/results/historical-al-curves/v1/aw3_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_attnwarm_code/epoch_3_step_24968,24968,code,3.4568350781368533 diff --git a/results/historical-al-curves/v1/aw4_code/al_curve.csv b/results/historical-al-curves/v1/aw4_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..048d1a744cead4c61cdbe46df89419788d7e962e --- /dev/null +++ b/results/historical-al-curves/v1/aw4_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_attnwarm_code/epoch_4_step_31210,31210,code,3.4458560193587418 diff --git a/results/historical-al-curves/v1/aw5_code/al_curve.csv b/results/historical-al-curves/v1/aw5_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..73b69a7bf4eff5234842464920d0f3abe9c3e42f --- /dev/null +++ b/results/historical-al-curves/v1/aw5_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_attnwarm_code/epoch_5_step_37452,37452,code,3.4715831174767637 diff --git a/results/historical-al-curves/v1/aw6_code/al_curve.csv b/results/historical-al-curves/v1/aw6_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a9427a52d38d5f70758a007694ca75bc6fd012e4 --- /dev/null +++ b/results/historical-al-curves/v1/aw6_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_attnwarm_code/epoch_6_step_43694,43694,code,3.471318656204769 diff --git a/results/historical-al-curves/v1/aw7_code/al_curve.csv b/results/historical-al-curves/v1/aw7_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c2b18c6dda13a4defd5f3c5d15213eb8a65683e7 --- /dev/null +++ b/results/historical-al-curves/v1/aw7_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_attnwarm_code/epoch_7_step_49936,49936,code,3.4620878285974777 diff --git a/results/historical-al-curves/v1/b0_code/al_curve.csv b/results/historical-al-curves/v1/b0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..f34b23a8debb0f22c86362e46e60cb1c96cf61c0 --- /dev/null +++ b/results/historical-al-curves/v1/b0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b0_experts89/code,0,code,3.476880817945979 diff --git a/results/historical-al-curves/v1/b0_creative_writing/al_curve.csv b/results/historical-al-curves/v1/b0_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..68f85d77e0f7f06c25e1e178416d30a04c79bee9 --- /dev/null +++ b/results/historical-al-curves/v1/b0_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b0_experts89/creative_writing,0,creative_writing,2.80445499360964 diff --git a/results/historical-al-curves/v1/b0_factual_qa/al_curve.csv b/results/historical-al-curves/v1/b0_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e1fd9bec673ceedd426c566db2a56d740f4a3a2a --- /dev/null +++ b/results/historical-al-curves/v1/b0_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b0_experts89/factual_qa,0,factual_qa,2.9171034987199804 diff --git a/results/historical-al-curves/v1/b0_general/al_curve.csv b/results/historical-al-curves/v1/b0_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..3c4950978c4a053980547fcf95acdf530af49463 --- /dev/null +++ b/results/historical-al-curves/v1/b0_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b0_experts89/general,0,general,3.1898075293547996 diff --git a/results/historical-al-curves/v1/b0_math/al_curve.csv b/results/historical-al-curves/v1/b0_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..fc43179e31594fcde4010d5f4ffc9bb1b507f0b1 --- /dev/null +++ b/results/historical-al-curves/v1/b0_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b0_experts89/math,0,math,5.07504956103087 diff --git a/results/historical-al-curves/v1/b1_code/al_curve.csv b/results/historical-al-curves/v1/b1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..98ec18f581f25372ee7b13ffeddd1e075b47e4b4 --- /dev/null +++ b/results/historical-al-curves/v1/b1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b1_experts89/code,0,code,3.585772741580107 diff --git a/results/historical-al-curves/v1/b1_creative_writing/al_curve.csv b/results/historical-al-curves/v1/b1_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..aeb5e94030a40cb423388ff5ef76cce7e07bce12 --- /dev/null +++ b/results/historical-al-curves/v1/b1_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b1_experts89/creative_writing,0,creative_writing,2.8997545780630545 diff --git a/results/historical-al-curves/v1/b1_factual_qa/al_curve.csv b/results/historical-al-curves/v1/b1_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..04718042360e0979027be88d87efdcc5a1a4b7e5 --- /dev/null +++ b/results/historical-al-curves/v1/b1_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b1_experts89/factual_qa,0,factual_qa,3.0083354382419802 diff --git a/results/historical-al-curves/v1/b1_general/al_curve.csv b/results/historical-al-curves/v1/b1_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..77067e52a945758adf1264d11b82648b8933f5f2 --- /dev/null +++ b/results/historical-al-curves/v1/b1_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b1_experts89/general,0,general,3.201011682524389 diff --git a/results/historical-al-curves/v1/b1_math/al_curve.csv b/results/historical-al-curves/v1/b1_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..efa2c97b2f4bbe451f8207be782494901ab24b78 --- /dev/null +++ b/results/historical-al-curves/v1/b1_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b1_experts89/math,0,math,5.245901639344262 diff --git a/results/historical-al-curves/v1/b2_code/al_curve.csv b/results/historical-al-curves/v1/b2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..712a20a53f20659d7513ce4f4d87964ac9ffc403 --- /dev/null +++ b/results/historical-al-curves/v1/b2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_experts89/code,0,code,3.6260046152621945 diff --git a/results/historical-al-curves/v1/b2_creative_writing/al_curve.csv b/results/historical-al-curves/v1/b2_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..efc0852d3706277ab509335a3b8c2ffe9368f0d7 --- /dev/null +++ b/results/historical-al-curves/v1/b2_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_experts89/creative_writing,0,creative_writing,2.917933130699088 diff --git a/results/historical-al-curves/v1/b2_factual_qa/al_curve.csv b/results/historical-al-curves/v1/b2_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..63a915ff0c0ccebf9a4518fbea91f28090527f42 --- /dev/null +++ b/results/historical-al-curves/v1/b2_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_experts89/factual_qa,0,factual_qa,2.9848351923799976 diff --git a/results/historical-al-curves/v1/b2_general/al_curve.csv b/results/historical-al-curves/v1/b2_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6e71847bceee266cd1119eeb77114c8a9013b146 --- /dev/null +++ b/results/historical-al-curves/v1/b2_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_experts89/general,0,general,3.2679955703211516 diff --git a/results/historical-al-curves/v1/b2_math/al_curve.csv b/results/historical-al-curves/v1/b2_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..d522f4374838b6456c34735254fcc71beca293be --- /dev/null +++ b/results/historical-al-curves/v1/b2_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_experts89/math,0,math,5.384615384615385 diff --git a/results/historical-al-curves/v1/b2_v3_cont_curve_8x5x1x6/al_curve.csv b/results/historical-al-curves/v1/b2_v3_cont_curve_8x5x1x6/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..521288e65fd0ebc04ec7465b3a09deadf64109de --- /dev/null +++ b/results/historical-al-curves/v1/b2_v3_cont_curve_8x5x1x6/al_curve.csv @@ -0,0 +1,7 @@ +ckpt,step,domain,accept_length +b2_v3_cont/epoch_0_step_7810,7810,code,3.0154247476465916 +b2_v3_cont/epoch_0_step_7810,7810,math,3.8799355816597196 +b2_v3_cont/epoch_1_step_15620,15620,code,3.0322748822887307 +b2_v3_cont/epoch_1_step_15620,15620,math,3.8816270826620163 +b2_v3_cont/epoch_2_step_23436,23436,code,3.0178032882576127 +b2_v3_cont/epoch_2_step_23436,23436,math,3.8764836163325573 diff --git a/results/historical-al-curves/v1/b2_v3_curve_8x5x1x6/al_curve.csv b/results/historical-al-curves/v1/b2_v3_curve_8x5x1x6/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..771ccc7c49cf1f69cd9a487bf1c0b50536b9cebb --- /dev/null +++ b/results/historical-al-curves/v1/b2_v3_curve_8x5x1x6/al_curve.csv @@ -0,0 +1,16 @@ +ckpt,step,domain,accept_length +b2_v3/epoch_0_step_7812,7812,code,2.8953517275664926 +b2_v3/epoch_0_step_7812,7812,math,3.7750105996644976 +b2_v3/epoch_1_step_15624,15624,code,2.977261461214961 +b2_v3/epoch_1_step_15624,15624,math,3.879274091193241 +b2_v3/epoch_2_step_23436,23436,code,3.019719040134809 +b2_v3/epoch_2_step_23436,23436,math,3.904198123999085 +b2_v3/epoch_4_step_39060,39060,code,3.0404264698679118 +b2_v3/epoch_4_step_39060,39060,math,3.900182836247286 +b2_v3/epoch_5_step_46872,46872,code,3.0385261270074184 +b2_v3/epoch_5_step_46872,46872,math,3.890919800117801 +b2_v3/epoch_3_step_31248,31248,math,3.9215164169955172 +b2_v3/epoch_3_step_31248,31248,code,3.0228667944262186 +b2_v3/epoch_3_step_31248,31248,commonsense,2.815646882294098 +b2_v3/epoch_3_step_31248,31248,finance,2.680979185757298 +b2_v3/epoch_3_step_31248,31248,chat,2.792580008768084 diff --git a/results/historical-al-curves/v1/b3_code/al_curve.csv b/results/historical-al-curves/v1/b3_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..34fad7a08d47051b1418bb1e2fbf4148126da059 --- /dev/null +++ b/results/historical-al-curves/v1/b3_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b3_experts89/code,0,code,3.6565559300272827 diff --git a/results/historical-al-curves/v1/b3_creative_writing/al_curve.csv b/results/historical-al-curves/v1/b3_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..5fbd663e337ac785df0f80b472d06f2ca25a1edf --- /dev/null +++ b/results/historical-al-curves/v1/b3_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b3_experts89/creative_writing,0,creative_writing,2.923486867148839 diff --git a/results/historical-al-curves/v1/b3_factual_qa/al_curve.csv b/results/historical-al-curves/v1/b3_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..67d7fa3a1979240e39267d7292eded4f95dbc4d3 --- /dev/null +++ b/results/historical-al-curves/v1/b3_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b3_experts89/factual_qa,0,factual_qa,3.012906491205871 diff --git a/results/historical-al-curves/v1/b3_general/al_curve.csv b/results/historical-al-curves/v1/b3_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a9c16588521851ec27cea445afcde7c82a2d5191 --- /dev/null +++ b/results/historical-al-curves/v1/b3_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b3_experts89/general,0,general,3.2509784735812133 diff --git a/results/historical-al-curves/v1/b3_math/al_curve.csv b/results/historical-al-curves/v1/b3_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..8b78925178939d4cdc3fa1e3c7c17e8acd305ca0 --- /dev/null +++ b/results/historical-al-curves/v1/b3_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b3_experts89/math,0,math,5.3862338443041775 diff --git a/results/historical-al-curves/v1/b4_code/al_curve.csv b/results/historical-al-curves/v1/b4_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..94d7a0a5a4c875a1561f4c9db04ca90ae6571229 --- /dev/null +++ b/results/historical-al-curves/v1/b4_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b4_experts89/code,0,code,3.7089370014650824 diff --git a/results/historical-al-curves/v1/b4_creative_writing/al_curve.csv b/results/historical-al-curves/v1/b4_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..aad19cf61fcdc873a5af7150a5536510f77cf234 --- /dev/null +++ b/results/historical-al-curves/v1/b4_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b4_experts89/creative_writing,0,creative_writing,2.9889083479276124 diff --git a/results/historical-al-curves/v1/b4_factual_qa/al_curve.csv b/results/historical-al-curves/v1/b4_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..1dd4f3bd006eccf8a641dddf13acb9c6a3a6ec86 --- /dev/null +++ b/results/historical-al-curves/v1/b4_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b4_experts89/factual_qa,0,factual_qa,3.024517276422764 diff --git a/results/historical-al-curves/v1/b4_general/al_curve.csv b/results/historical-al-curves/v1/b4_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..31a07de27526db2425bdc1e1614675f62105f40b --- /dev/null +++ b/results/historical-al-curves/v1/b4_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b4_experts89/general,0,general,3.277262037380864 diff --git a/results/historical-al-curves/v1/b4_math/al_curve.csv b/results/historical-al-curves/v1/b4_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..ee45f81c239b29e215db08b44aff05c691e98ccb --- /dev/null +++ b/results/historical-al-curves/v1/b4_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b4_experts89/math,0,math,5.435244161358811 diff --git a/results/historical-al-curves/v1/b5_code/al_curve.csv b/results/historical-al-curves/v1/b5_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a97217b83e88cd0e7a20a9fe20d14aab0a38b323 --- /dev/null +++ b/results/historical-al-curves/v1/b5_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/code,0,code,3.718622490615309 diff --git a/results/historical-al-curves/v1/b5_creative_writing/al_curve.csv b/results/historical-al-curves/v1/b5_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e3a1212594bf9e6e9b18306fe081ed586a130ee8 --- /dev/null +++ b/results/historical-al-curves/v1/b5_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/creative_writing,0,creative_writing,2.9755908562572646 diff --git a/results/historical-al-curves/v1/b5_factual_qa/al_curve.csv b/results/historical-al-curves/v1/b5_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..70f195e0344698abb76151fc23f79e0fa4662ec5 --- /dev/null +++ b/results/historical-al-curves/v1/b5_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/factual_qa,0,factual_qa,3.0264563724243194 diff --git a/results/historical-al-curves/v1/b5_general/al_curve.csv b/results/historical-al-curves/v1/b5_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..997720d82fb8a0b653a19e6becdeae234ccdd593 --- /dev/null +++ b/results/historical-al-curves/v1/b5_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/general,0,general,3.258826393961529 diff --git a/results/historical-al-curves/v1/b5_math/al_curve.csv b/results/historical-al-curves/v1/b5_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a23dfaa5d5007c5ca2e97e0d62d73b1676f35498 --- /dev/null +++ b/results/historical-al-curves/v1/b5_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/math,0,math,5.47677261613692 diff --git a/results/historical-al-curves/v1/b6_code/al_curve.csv b/results/historical-al-curves/v1/b6_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..d5eb916a2c3aa1b2baac300c74fb147f03d7a07d --- /dev/null +++ b/results/historical-al-curves/v1/b6_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b6_experts89/code,0,code,3.715893337682459 diff --git a/results/historical-al-curves/v1/b6_creative_writing/al_curve.csv b/results/historical-al-curves/v1/b6_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..466f20ffc8ddacd662725d505abc9c6c4e49c63a --- /dev/null +++ b/results/historical-al-curves/v1/b6_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b6_experts89/creative_writing,0,creative_writing,2.9732868757259 diff --git a/results/historical-al-curves/v1/b6_factual_qa/al_curve.csv b/results/historical-al-curves/v1/b6_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c673d3367200bf32ee24cdeb921ed4adbad24dbb --- /dev/null +++ b/results/historical-al-curves/v1/b6_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b6_experts89/factual_qa,0,factual_qa,3.024399542508578 diff --git a/results/historical-al-curves/v1/b6_general/al_curve.csv b/results/historical-al-curves/v1/b6_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..58a90c87af68cb9d8c39ed4a3cfb99ad18ed6303 --- /dev/null +++ b/results/historical-al-curves/v1/b6_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b6_experts89/general,0,general,3.257661191468497 diff --git a/results/historical-al-curves/v1/b6_math/al_curve.csv b/results/historical-al-curves/v1/b6_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a2425ab779061ee4f091764bc7f223a65fc1ff06 --- /dev/null +++ b/results/historical-al-curves/v1/b6_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b6_experts89/math,0,math,5.480122324159021 diff --git a/results/historical-al-curves/v1/b7_code/al_curve.csv b/results/historical-al-curves/v1/b7_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..18a40b879b5751fefb1b7bf20d6fc66529f64705 --- /dev/null +++ b/results/historical-al-curves/v1/b7_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b7_experts89/code,0,code,3.7098428722624766 diff --git a/results/historical-al-curves/v1/b7_creative_writing/al_curve.csv b/results/historical-al-curves/v1/b7_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..06c11d49138c5fc8e46a2cb62ce373a1fcba2caa --- /dev/null +++ b/results/historical-al-curves/v1/b7_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b7_experts89/creative_writing,0,creative_writing,2.9618202853837254 diff --git a/results/historical-al-curves/v1/b7_factual_qa/al_curve.csv b/results/historical-al-curves/v1/b7_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..55060ef42ff225f459d15bddc305117500cee66a --- /dev/null +++ b/results/historical-al-curves/v1/b7_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b7_experts89/factual_qa,0,factual_qa,3.080509024801974 diff --git a/results/historical-al-curves/v1/b7_general/al_curve.csv b/results/historical-al-curves/v1/b7_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c44b0f79e0cc664c211ef7cee9653a73054b5e9c --- /dev/null +++ b/results/historical-al-curves/v1/b7_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b7_experts89/general,0,general,3.2750061440157285 diff --git a/results/historical-al-curves/v1/b7_math/al_curve.csv b/results/historical-al-curves/v1/b7_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2fe4ea6638bc2ad118728b68833277010aadeb46 --- /dev/null +++ b/results/historical-al-curves/v1/b7_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b7_experts89/math,0,math,5.458422174840085 diff --git a/results/historical-al-curves/v1/c0_code/al_curve.csv b/results/historical-al-curves/v1/c0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..8e614012efee175c43f034f0cecf13478e44fcc4 --- /dev/null +++ b/results/historical-al-curves/v1/c0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c0_experts89/code,0,code,3.185459629500175 diff --git a/results/historical-al-curves/v1/c0_creative_writing/al_curve.csv b/results/historical-al-curves/v1/c0_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..7194dfa58d3e63e8732d47f9cf3ae6d0f2fa7e73 --- /dev/null +++ b/results/historical-al-curves/v1/c0_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c0_experts89/creative_writing,0,creative_writing,2.5954714430550863 diff --git a/results/historical-al-curves/v1/c0_factual_qa/al_curve.csv b/results/historical-al-curves/v1/c0_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2055ea24a027215da022397adf337b7226239709 --- /dev/null +++ b/results/historical-al-curves/v1/c0_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c0_experts89/factual_qa,0,factual_qa,2.6727353959566624 diff --git a/results/historical-al-curves/v1/c0_general/al_curve.csv b/results/historical-al-curves/v1/c0_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..df9fe16775a44feb0eadaf78e92a35cb85bffad0 --- /dev/null +++ b/results/historical-al-curves/v1/c0_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c0_experts89/general,0,general,2.88092901537455 diff --git a/results/historical-al-curves/v1/c0_math/al_curve.csv b/results/historical-al-curves/v1/c0_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..663b546ffb13f3edd94cc5477a94f566ec10ac59 --- /dev/null +++ b/results/historical-al-curves/v1/c0_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c0_experts89/math,0,math,4.626904208623806 diff --git a/results/historical-al-curves/v1/c1_code/al_curve.csv b/results/historical-al-curves/v1/c1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..64dc90e2380c43bbda062101340c33d46a78efdf --- /dev/null +++ b/results/historical-al-curves/v1/c1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c1_experts89/code,0,code,3.31524190614769 diff --git a/results/historical-al-curves/v1/c1_creative_writing/al_curve.csv b/results/historical-al-curves/v1/c1_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..16041207f1c513f598b9925313ff3fd6277f196b --- /dev/null +++ b/results/historical-al-curves/v1/c1_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c1_experts89/creative_writing,0,creative_writing,2.6787582839204744 diff --git a/results/historical-al-curves/v1/c1_factual_qa/al_curve.csv b/results/historical-al-curves/v1/c1_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..107b0101d41f407ed8fd895ba30b7f887c5e7d5b --- /dev/null +++ b/results/historical-al-curves/v1/c1_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c1_experts89/factual_qa,0,factual_qa,2.753125 diff --git a/results/historical-al-curves/v1/c1_general/al_curve.csv b/results/historical-al-curves/v1/c1_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c53af17a18b891e8f1342c6d1a8282993edfa537 --- /dev/null +++ b/results/historical-al-curves/v1/c1_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c1_experts89/general,0,general,2.9550673437845 diff --git a/results/historical-al-curves/v1/c1_math/al_curve.csv b/results/historical-al-curves/v1/c1_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6a05cdae634de7eaa1b3816e4fa9483b95caf624 --- /dev/null +++ b/results/historical-al-curves/v1/c1_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c1_experts89/math,0,math,4.828887092427917 diff --git a/results/historical-al-curves/v1/c2_code/al_curve.csv b/results/historical-al-curves/v1/c2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6087ca66b61f9fdbe3b9941c7be49f752a40a2fc --- /dev/null +++ b/results/historical-al-curves/v1/c2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c2_experts89/code,0,code,3.394517282479142 diff --git a/results/historical-al-curves/v1/c2_creative_writing/al_curve.csv b/results/historical-al-curves/v1/c2_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..fa17060533ccfa699a0dd4b3d50b8ea8d5bd792e --- /dev/null +++ b/results/historical-al-curves/v1/c2_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c2_experts89/creative_writing,0,creative_writing,2.7467811158798283 diff --git a/results/historical-al-curves/v1/c2_factual_qa/al_curve.csv b/results/historical-al-curves/v1/c2_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..be5e80809f13060543dc00301016f319c0281c4d --- /dev/null +++ b/results/historical-al-curves/v1/c2_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c2_experts89/factual_qa,0,factual_qa,2.797416324133881 diff --git a/results/historical-al-curves/v1/c2_general/al_curve.csv b/results/historical-al-curves/v1/c2_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e1086f17c85a33f6cec4fbf4f36d975566431615 --- /dev/null +++ b/results/historical-al-curves/v1/c2_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c2_experts89/general,0,general,3.011331444759207 diff --git a/results/historical-al-curves/v1/c2_math/al_curve.csv b/results/historical-al-curves/v1/c2_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..4db1307d64bb68d1e22b886656e630762f93ede9 --- /dev/null +++ b/results/historical-al-curves/v1/c2_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c2_experts89/math,0,math,4.984700973574409 diff --git a/results/historical-al-curves/v1/c3_code/al_curve.csv b/results/historical-al-curves/v1/c3_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..acceb0f27ce1971e6bbc2f0a16892d7752c86e0a --- /dev/null +++ b/results/historical-al-curves/v1/c3_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c3_experts89/code,0,code,3.419737335834897 diff --git a/results/historical-al-curves/v1/c3_creative_writing/al_curve.csv b/results/historical-al-curves/v1/c3_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2efff35500007d476f8baafa90df74ee83cb3b21 --- /dev/null +++ b/results/historical-al-curves/v1/c3_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c3_experts89/creative_writing,0,creative_writing,2.751209027404621 diff --git a/results/historical-al-curves/v1/c3_factual_qa/al_curve.csv b/results/historical-al-curves/v1/c3_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..b967b183b1d5383b9a54db6a5d03488ff9e438b0 --- /dev/null +++ b/results/historical-al-curves/v1/c3_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c3_experts89/factual_qa,0,factual_qa,2.7999293036408623 diff --git a/results/historical-al-curves/v1/c3_general/al_curve.csv b/results/historical-al-curves/v1/c3_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..24f5879e875a21af34d487dff285735dce50d043 --- /dev/null +++ b/results/historical-al-curves/v1/c3_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c3_experts89/general,0,general,3.02803738317757 diff --git a/results/historical-al-curves/v1/c3_math/al_curve.csv b/results/historical-al-curves/v1/c3_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..767981ad4c13fe51a1abce0d54f56635e2a066b2 --- /dev/null +++ b/results/historical-al-curves/v1/c3_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c3_experts89/math,0,math,5.063577281718 diff --git a/results/historical-al-curves/v1/c4_code/al_curve.csv b/results/historical-al-curves/v1/c4_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e34750fd6ff2ec1f6b13fc3c518dfc73ff2b1999 --- /dev/null +++ b/results/historical-al-curves/v1/c4_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c4_experts89/code,0,code,3.4328762995329214 diff --git a/results/historical-al-curves/v1/c4_creative_writing/al_curve.csv b/results/historical-al-curves/v1/c4_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..d9fe14985d68164885f1a1e679c7fa7d95cdb010 --- /dev/null +++ b/results/historical-al-curves/v1/c4_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c4_experts89/creative_writing,0,creative_writing,2.7635840230298667 diff --git a/results/historical-al-curves/v1/c4_factual_qa/al_curve.csv b/results/historical-al-curves/v1/c4_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..d4b5d06256071a7078077c4bd3283f1996a0978b --- /dev/null +++ b/results/historical-al-curves/v1/c4_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c4_experts89/factual_qa,0,factual_qa,2.810434372049103 diff --git a/results/historical-al-curves/v1/c4_general/al_curve.csv b/results/historical-al-curves/v1/c4_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c89370d115d6262f088216a8ae9a166909b54f53 --- /dev/null +++ b/results/historical-al-curves/v1/c4_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c4_experts89/general,0,general,3.047881501894592 diff --git a/results/historical-al-curves/v1/c4_math/al_curve.csv b/results/historical-al-curves/v1/c4_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6e88173d09207fedf7b55df6b748c3c71fd643a0 --- /dev/null +++ b/results/historical-al-curves/v1/c4_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c4_experts89/math,0,math,5.112696148359486 diff --git a/results/historical-al-curves/v1/c5_code/al_curve.csv b/results/historical-al-curves/v1/c5_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..156bfb4ce8565a07dc05e0eedb5e18c280da73f1 --- /dev/null +++ b/results/historical-al-curves/v1/c5_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c5_experts89/code,0,code,3.4396135265700485 diff --git a/results/historical-al-curves/v1/c5_creative_writing/al_curve.csv b/results/historical-al-curves/v1/c5_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..8d633478cc115d9d845e7753ab9aea70ce16599c --- /dev/null +++ b/results/historical-al-curves/v1/c5_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c5_experts89/creative_writing,0,creative_writing,2.76657060518732 diff --git a/results/historical-al-curves/v1/c5_factual_qa/al_curve.csv b/results/historical-al-curves/v1/c5_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..ccd20fa3e9b8f7b541ba2c41616c7fc721ec0c63 --- /dev/null +++ b/results/historical-al-curves/v1/c5_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c5_experts89/factual_qa,0,factual_qa,2.8178156834873254 diff --git a/results/historical-al-curves/v1/c5_general/al_curve.csv b/results/historical-al-curves/v1/c5_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..59741cdceee5ecfbe51a402671cf51fa912555de --- /dev/null +++ b/results/historical-al-curves/v1/c5_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c5_experts89/general,0,general,3.059820193637621 diff --git a/results/historical-al-curves/v1/c5_math/al_curve.csv b/results/historical-al-curves/v1/c5_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c8620ffd438b7eb8e0f17a608bbb8085b8c1387b --- /dev/null +++ b/results/historical-al-curves/v1/c5_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c5_experts89/math,0,math,5.147945992530882 diff --git a/results/historical-al-curves/v1/c6_code/al_curve.csv b/results/historical-al-curves/v1/c6_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c886c2251132801f6cb4e593458c5c142523ed55 --- /dev/null +++ b/results/historical-al-curves/v1/c6_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c6_experts89/code,0,code,3.461298898594759 diff --git a/results/historical-al-curves/v1/c6_creative_writing/al_curve.csv b/results/historical-al-curves/v1/c6_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..f2225a3ad5a1352475074486d8e08b855b5dee9c --- /dev/null +++ b/results/historical-al-curves/v1/c6_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c6_experts89/creative_writing,0,creative_writing,2.763086886130599 diff --git a/results/historical-al-curves/v1/c6_factual_qa/al_curve.csv b/results/historical-al-curves/v1/c6_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2c41d853086a85522be5219f32bd160a480923eb --- /dev/null +++ b/results/historical-al-curves/v1/c6_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c6_experts89/factual_qa,0,factual_qa,2.814132104454685 diff --git a/results/historical-al-curves/v1/c6_general/al_curve.csv b/results/historical-al-curves/v1/c6_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..07357458e8fab4ed71f88bb09c92f0f4f2a08b08 --- /dev/null +++ b/results/historical-al-curves/v1/c6_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c6_experts89/general,0,general,3.0381609195402297 diff --git a/results/historical-al-curves/v1/c6_math/al_curve.csv b/results/historical-al-curves/v1/c6_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..526bc15d72d2d62ab6d6a0e8973984b1d8c5cb7c --- /dev/null +++ b/results/historical-al-curves/v1/c6_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c6_experts89/math,0,math,5.147945992530882 diff --git a/results/historical-al-curves/v1/c7_code/al_curve.csv b/results/historical-al-curves/v1/c7_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2fa860fad8899f0f30301ee4387eebe0c9607ae4 --- /dev/null +++ b/results/historical-al-curves/v1/c7_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c7_experts89/code,0,code,3.472641365645481 diff --git a/results/historical-al-curves/v1/c7_creative_writing/al_curve.csv b/results/historical-al-curves/v1/c7_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..bca3c96664d322e6fe75e8b5e673e2af361546f5 --- /dev/null +++ b/results/historical-al-curves/v1/c7_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c7_experts89/creative_writing,0,creative_writing,2.7660723933009184 diff --git a/results/historical-al-curves/v1/c7_factual_qa/al_curve.csv b/results/historical-al-curves/v1/c7_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..16850c57ae4bbc6b90c189f0552e26d68cb474a6 --- /dev/null +++ b/results/historical-al-curves/v1/c7_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c7_experts89/factual_qa,0,factual_qa,2.8272176701104383 diff --git a/results/historical-al-curves/v1/c7_general/al_curve.csv b/results/historical-al-curves/v1/c7_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..0ba907ef0ba8bbfbc303660f8e5bc5e100a575b5 --- /dev/null +++ b/results/historical-al-curves/v1/c7_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c7_experts89/general,0,general,3.0346008907159985 diff --git a/results/historical-al-curves/v1/c7_math/al_curve.csv b/results/historical-al-curves/v1/c7_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..8325e18db9d36e7d30423bbdaaa1eb506f87e5c3 --- /dev/null +++ b/results/historical-al-curves/v1/c7_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +c7_experts89/math,0,math,5.1361421610776725 diff --git a/results/historical-al-curves/v1/d0_code/al_curve.csv b/results/historical-al-curves/v1/d0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..f03128ba18e9901dd3302f617708ee391c30a81e --- /dev/null +++ b/results/historical-al-curves/v1/d0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d0_experts89/code,0,code,3.4705255140898705 diff --git a/results/historical-al-curves/v1/d0_code_bench/al_curve.csv b/results/historical-al-curves/v1/d0_code_bench/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6832cbf1553fa6e96ba50ae89b00f3ce6f743b8e --- /dev/null +++ b/results/historical-al-curves/v1/d0_code_bench/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_v3/dflash-D0,0,code,2.436921760521953 diff --git a/results/historical-al-curves/v1/d0_creative_writing/al_curve.csv b/results/historical-al-curves/v1/d0_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..50179375ba236b2fbc579f82a157721f52c8b1ad --- /dev/null +++ b/results/historical-al-curves/v1/d0_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d0_experts89/creative_writing,0,creative_writing,2.776070847641424 diff --git a/results/historical-al-curves/v1/d0_creative_writing_bench/al_curve.csv b/results/historical-al-curves/v1/d0_creative_writing_bench/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..84ee8fa606633c58817795486dfc864d8fc71061 --- /dev/null +++ b/results/historical-al-curves/v1/d0_creative_writing_bench/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_v3/dflash-D0,0,creative_writing,2.1061291649526943 diff --git a/results/historical-al-curves/v1/d0_factual_qa/al_curve.csv b/results/historical-al-curves/v1/d0_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..38f512d3bf7fc8c2a90d5636fa65ab7d25a01abc --- /dev/null +++ b/results/historical-al-curves/v1/d0_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d0_experts89/factual_qa,0,factual_qa,2.9147971067794534 diff --git a/results/historical-al-curves/v1/d0_factual_qa_bench/al_curve.csv b/results/historical-al-curves/v1/d0_factual_qa_bench/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..612ac89eea0c16504f828b0b9a99c911a0cebeef --- /dev/null +++ b/results/historical-al-curves/v1/d0_factual_qa_bench/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_v3/dflash-D0,0,factual_qa,2.232189973614776 diff --git a/results/historical-al-curves/v1/d0_general/al_curve.csv b/results/historical-al-curves/v1/d0_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..cd8a0058bc2e9b908062dde2e9ec7c4018ccec87 --- /dev/null +++ b/results/historical-al-curves/v1/d0_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d0_experts89/general,0,general,3.1655237644833294 diff --git a/results/historical-al-curves/v1/d0_general_bench/al_curve.csv b/results/historical-al-curves/v1/d0_general_bench/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..731d43b337d607d2da21373be9b509ea59b09f9d --- /dev/null +++ b/results/historical-al-curves/v1/d0_general_bench/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_v3/dflash-D0,0,general,2.4312111233077203 diff --git a/results/historical-al-curves/v1/d0_math/al_curve.csv b/results/historical-al-curves/v1/d0_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..0adbe2d9b1aaf1b346f655e0688e73e3c901a7a4 --- /dev/null +++ b/results/historical-al-curves/v1/d0_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d0_experts89/math,0,math,5.07361268403171 diff --git a/results/historical-al-curves/v1/d0_math_bench/al_curve.csv b/results/historical-al-curves/v1/d0_math_bench/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..102126bc5951af449126c0df407adcbcdd2a4256 --- /dev/null +++ b/results/historical-al-curves/v1/d0_math_bench/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b2_v3/dflash-D0,0,math,3.3773087071240107 diff --git a/results/historical-al-curves/v1/d1_code/al_curve.csv b/results/historical-al-curves/v1/d1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..0f10699efbd0326084d069e64ec9b4c6f8639a81 --- /dev/null +++ b/results/historical-al-curves/v1/d1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d1_experts89/code,0,code,3.535966477845891 diff --git a/results/historical-al-curves/v1/d1_creative_writing/al_curve.csv b/results/historical-al-curves/v1/d1_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..d50105527b2f18652ababea2c097b65e9f3d1018 --- /dev/null +++ b/results/historical-al-curves/v1/d1_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d1_experts89/creative_writing,0,creative_writing,2.8704914969164643 diff --git a/results/historical-al-curves/v1/d1_factual_qa/al_curve.csv b/results/historical-al-curves/v1/d1_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..3049a4305ada59eec30a17bd5f3fd657cbefebc7 --- /dev/null +++ b/results/historical-al-curves/v1/d1_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d1_experts89/factual_qa,0,factual_qa,3.000252940432528 diff --git a/results/historical-al-curves/v1/d1_general/al_curve.csv b/results/historical-al-curves/v1/d1_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..50dc5d05a785008ba74de4429c2bed440c7cb5b1 --- /dev/null +++ b/results/historical-al-curves/v1/d1_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d1_experts89/general,0,general,3.2252559726962455 diff --git a/results/historical-al-curves/v1/d1_math/al_curve.csv b/results/historical-al-curves/v1/d1_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..42606fe71b8b6df7e5fba087340d2a0a57aaedd2 --- /dev/null +++ b/results/historical-al-curves/v1/d1_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d1_experts89/math,0,math,5.173210161662817 diff --git a/results/historical-al-curves/v1/d2_code/al_curve.csv b/results/historical-al-curves/v1/d2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..ff2b9ec8421852edfd4a1b289fcbe604ef8cca84 --- /dev/null +++ b/results/historical-al-curves/v1/d2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d2_experts89/code,0,code,3.553337492202121 diff --git a/results/historical-al-curves/v1/d2_creative_writing/al_curve.csv b/results/historical-al-curves/v1/d2_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..d5807e98bd0a9093bea884f99a3a0f6552bf0b88 --- /dev/null +++ b/results/historical-al-curves/v1/d2_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d2_experts89/creative_writing,0,creative_writing,2.8753275926619244 diff --git a/results/historical-al-curves/v1/d2_factual_qa/al_curve.csv b/results/historical-al-curves/v1/d2_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..7bb6323abc90495d6739ccacaf13b44def875ece --- /dev/null +++ b/results/historical-al-curves/v1/d2_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d2_experts89/factual_qa,0,factual_qa,2.9420630018529956 diff --git a/results/historical-al-curves/v1/d2_general/al_curve.csv b/results/historical-al-curves/v1/d2_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a2f34fa796675c14540a7b7d374f9e6103cf8f7d --- /dev/null +++ b/results/historical-al-curves/v1/d2_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d2_experts89/general,0,general,3.2306565779838907 diff --git a/results/historical-al-curves/v1/d2_math/al_curve.csv b/results/historical-al-curves/v1/d2_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..445aff7b404f86bf5cb3bed5d1d7242409710997 --- /dev/null +++ b/results/historical-al-curves/v1/d2_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d2_experts89/math,0,math,5.256673511293634 diff --git a/results/historical-al-curves/v1/d3_code/al_curve.csv b/results/historical-al-curves/v1/d3_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..28c20e68d9d55c1fe82ec074df9ec11da20f831b --- /dev/null +++ b/results/historical-al-curves/v1/d3_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d3_experts89/code,0,code,3.6056338028169015 diff --git a/results/historical-al-curves/v1/d3_creative_writing/al_curve.csv b/results/historical-al-curves/v1/d3_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2f2646dde11cb85fb19237949f07011a866b7aa3 --- /dev/null +++ b/results/historical-al-curves/v1/d3_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d3_experts89/creative_writing,0,creative_writing,2.9173789173789175 diff --git a/results/historical-al-curves/v1/d3_factual_qa/al_curve.csv b/results/historical-al-curves/v1/d3_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..0fc76e2bd19598456f2900b3d18f149af19f3f90 --- /dev/null +++ b/results/historical-al-curves/v1/d3_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d3_experts89/factual_qa,0,factual_qa,3.0036736762097798 diff --git a/results/historical-al-curves/v1/d3_general/al_curve.csv b/results/historical-al-curves/v1/d3_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..70eb926c9dabcbf7a33a954c2df90dc57327ff50 --- /dev/null +++ b/results/historical-al-curves/v1/d3_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d3_experts89/general,0,general,3.204726308174584 diff --git a/results/historical-al-curves/v1/d3_math/al_curve.csv b/results/historical-al-curves/v1/d3_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..b0e1361794f68f97a6b11169f336ba99fe864ecb --- /dev/null +++ b/results/historical-al-curves/v1/d3_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d3_experts89/math,0,math,5.286135693215339 diff --git a/results/historical-al-curves/v1/d4_code/al_curve.csv b/results/historical-al-curves/v1/d4_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..7e112ed3833c402f9876941820736f43fed4879a --- /dev/null +++ b/results/historical-al-curves/v1/d4_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d4_experts89/code,0,code,3.628603280777194 diff --git a/results/historical-al-curves/v1/d4_creative_writing/al_curve.csv b/results/historical-al-curves/v1/d4_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..539b66d3c3142bb1519e72863a9010f05773ef7d --- /dev/null +++ b/results/historical-al-curves/v1/d4_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d4_experts89/creative_writing,0,creative_writing,2.866206381787647 diff --git a/results/historical-al-curves/v1/d5_code/al_curve.csv b/results/historical-al-curves/v1/d5_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c902921a4ddd913dd0b682d8be9e658b454abbfb --- /dev/null +++ b/results/historical-al-curves/v1/d5_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d5_experts89/code,0,code,3.6110626832554087 diff --git a/results/historical-al-curves/v1/d5_creative_writing/al_curve.csv b/results/historical-al-curves/v1/d5_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..203cc346c2fd0afa46778c9c5e82bf5845d20e02 --- /dev/null +++ b/results/historical-al-curves/v1/d5_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d5_experts89/creative_writing,0,creative_writing,2.8764044943820224 diff --git a/results/historical-al-curves/v1/d5_factual_qa/al_curve.csv b/results/historical-al-curves/v1/d5_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..96f3a0aa6ed310a7d6140635f8cd322fb76c3611 --- /dev/null +++ b/results/historical-al-curves/v1/d5_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d5_experts89/factual_qa,0,factual_qa,2.944245271356163 diff --git a/results/historical-al-curves/v1/d5_general/al_curve.csv b/results/historical-al-curves/v1/d5_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..74a442fbc68ac5ae9197c1763df2685a733ffedc --- /dev/null +++ b/results/historical-al-curves/v1/d5_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d5_experts89/general,0,general,3.224644663936401 diff --git a/results/historical-al-curves/v1/d5_math/al_curve.csv b/results/historical-al-curves/v1/d5_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..90aeb51985c31a1995d8aea124cb225d1be42ac9 --- /dev/null +++ b/results/historical-al-curves/v1/d5_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d5_experts89/math,0,math,5.34446764091858 diff --git a/results/historical-al-curves/v1/d6_code/al_curve.csv b/results/historical-al-curves/v1/d6_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..645ea93ddaab543e9fd6a160708ef7dff8a887fb --- /dev/null +++ b/results/historical-al-curves/v1/d6_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d6_experts89/code,0,code,3.6480666079577295 diff --git a/results/historical-al-curves/v1/d6_creative_writing/al_curve.csv b/results/historical-al-curves/v1/d6_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..32ed7b5c0864204731e7a08dc8bfc4a9eede5aa6 --- /dev/null +++ b/results/historical-al-curves/v1/d6_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d6_experts89/creative_writing,0,creative_writing,2.905239266124456 diff --git a/results/historical-al-curves/v1/d6_factual_qa/al_curve.csv b/results/historical-al-curves/v1/d6_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..74db5fb92776208b9efc97f18fbf40fae4202fb8 --- /dev/null +++ b/results/historical-al-curves/v1/d6_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d6_experts89/factual_qa,0,factual_qa,3.062386746052291 diff --git a/results/historical-al-curves/v1/d6_general/al_curve.csv b/results/historical-al-curves/v1/d6_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6bd9e093f42a3d0f8cce2bb3627b5717da3d884f --- /dev/null +++ b/results/historical-al-curves/v1/d6_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d6_experts89/general,0,general,3.22952823270594 diff --git a/results/historical-al-curves/v1/d6_math/al_curve.csv b/results/historical-al-curves/v1/d6_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..73f5ae3a169ba1f84e694f056c605ae8fe5a2a56 --- /dev/null +++ b/results/historical-al-curves/v1/d6_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d6_experts89/math,0,math,5.320665083135392 diff --git a/results/historical-al-curves/v1/d7_code/al_curve.csv b/results/historical-al-curves/v1/d7_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..124f71aae4c24b2ae471b553ae28df16c9eeb313 --- /dev/null +++ b/results/historical-al-curves/v1/d7_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d7_experts89/code,0,code,3.625716104392107 diff --git a/results/historical-al-curves/v1/d7_creative_writing/al_curve.csv b/results/historical-al-curves/v1/d7_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2ce65fe46bd39c6988380a0e8d49ed5724e239cc --- /dev/null +++ b/results/historical-al-curves/v1/d7_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d7_experts89/creative_writing,0,creative_writing,2.890477982687241 diff --git a/results/historical-al-curves/v1/d7_factual_qa/al_curve.csv b/results/historical-al-curves/v1/d7_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..4d3f13b16a671a490e0b25e925c72efd1b64fdc5 --- /dev/null +++ b/results/historical-al-curves/v1/d7_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d7_experts89/factual_qa,0,factual_qa,2.9945932352571356 diff --git a/results/historical-al-curves/v1/d7_general/al_curve.csv b/results/historical-al-curves/v1/d7_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..72cd6d5d220f5595750d2ff674e3b9e18168fcd7 --- /dev/null +++ b/results/historical-al-curves/v1/d7_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d7_experts89/general,0,general,3.231192660550459 diff --git a/results/historical-al-curves/v1/d7_math/al_curve.csv b/results/historical-al-curves/v1/d7_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..fb1e710f481a1f51bec2ac9a2f4d4458640260ef --- /dev/null +++ b/results/historical-al-curves/v1/d7_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +d7_experts89/math,0,math,5.352449223416965 diff --git a/results/historical-al-curves/v1/gendense_l0/al_curve.csv b/results/historical-al-curves/v1/gendense_l0/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..5e97cdf6d5f039660f835238d1f6ddfecb1a5a59 --- /dev/null +++ b/results/historical-al-curves/v1/gendense_l0/al_curve.csv @@ -0,0 +1,8 @@ +ckpt,step,domain,accept_length +dflash_gen800k/step_1250,1250,code,2.821897448600446 +dflash_gen800k/step_11250,11250,code,3.2101444170482565 +dflash_gen800k/step_21250,21250,code,3.290821116487326 +dflash_gen800k/step_31250,31250,code,3.339048875210669 +dflash_gen800k/step_41250,41250,code,3.403390843229517 +dflash_gen800k/step_51250,51250,code,3.405680119581465 +dflash_gen800k/step_61250,61250,code,3.4089922944564974 diff --git a/results/historical-al-curves/v1/gendense_l1/al_curve.csv b/results/historical-al-curves/v1/gendense_l1/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..f30b16202bfac36d08949b4b784c5e760c50957e --- /dev/null +++ b/results/historical-al-curves/v1/gendense_l1/al_curve.csv @@ -0,0 +1,8 @@ +ckpt,step,domain,accept_length +dflash_gen800k/step_2500,2500,code,2.937028681920722 +dflash_gen800k/step_12500,12500,code,3.2017987633501965 +dflash_gen800k/step_22500,22500,code,3.3339186420836993 +dflash_gen800k/step_32500,32500,code,3.364441819255759 +dflash_gen800k/step_42500,42500,code,3.397807769741257 +dflash_gen800k/step_52500,52500,code,3.4207642068913744 +dflash_gen800k/step_62500,62500,code,3.3988215111508913 diff --git a/results/historical-al-curves/v1/gendense_l2/al_curve.csv b/results/historical-al-curves/v1/gendense_l2/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..cda93ccfe435a4194470f1cb8a6a0d014c9fdd9c --- /dev/null +++ b/results/historical-al-curves/v1/gendense_l2/al_curve.csv @@ -0,0 +1,8 @@ +ckpt,step,domain,accept_length +dflash_gen800k/step_3750,3750,code,3.0241571542341386 +dflash_gen800k/step_13750,13750,code,3.2451217775245693 +dflash_gen800k/step_23750,23750,code,3.34149739678815 +dflash_gen800k/step_33750,33750,code,3.375907541858053 +dflash_gen800k/step_43750,43750,code,3.4346875706640536 +dflash_gen800k/step_53750,53750,code,3.4241058010219416 +dflash_gen800k/step_63750,63750,code,3.4215347649797265 diff --git a/results/historical-al-curves/v1/gendense_l3/al_curve.csv b/results/historical-al-curves/v1/gendense_l3/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2aee2d7e105987e7f8277629044aea6c6237b9fd --- /dev/null +++ b/results/historical-al-curves/v1/gendense_l3/al_curve.csv @@ -0,0 +1,8 @@ +ckpt,step,domain,accept_length +dflash_gen800k/step_5000,5000,code,3.0496586802302237 +dflash_gen800k/step_15000,15000,code,3.246740292126826 +dflash_gen800k/step_25000,25000,code,3.34149739678815 +dflash_gen800k/step_35000,35000,code,3.3819207362327446 +dflash_gen800k/step_45000,45000,code,3.396794632873649 +dflash_gen800k/step_55000,55000,code,3.4235912847483094 +dflash_gen800k/step_65000,65000,code,3.4031366691560865 diff --git a/results/historical-al-curves/v1/gendense_l4/al_curve.csv b/results/historical-al-curves/v1/gendense_l4/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..846bcf29e918b2049acb30225c5f3a2bf14fe6c0 --- /dev/null +++ b/results/historical-al-curves/v1/gendense_l4/al_curve.csv @@ -0,0 +1,7 @@ +ckpt,step,domain,accept_length +dflash_gen800k/step_6250,6250,code,3.1062031356509885 +dflash_gen800k/step_16250,16250,code,3.2613799026624677 +dflash_gen800k/step_26250,26250,code,3.3400278531114855 +dflash_gen800k/step_36250,36250,code,3.3809170500074197 +dflash_gen800k/step_46250,46250,code,3.411033760011977 +dflash_gen800k/step_56250,56250,code,3.385438335809807 diff --git a/results/historical-al-curves/v1/gendense_l5/al_curve.csv b/results/historical-al-curves/v1/gendense_l5/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..70c4672684643f6e0c19fb0a61235506cab32eab --- /dev/null +++ b/results/historical-al-curves/v1/gendense_l5/al_curve.csv @@ -0,0 +1,7 @@ +ckpt,step,domain,accept_length +dflash_gen800k/step_7500,7500,code,3.1273076659117427 +dflash_gen800k/step_17500,17500,code,3.2714480580084717 +dflash_gen800k/step_27500,27500,code,3.336359642700249 +dflash_gen800k/step_37500,37500,code,3.3897195566465816 +dflash_gen800k/step_47500,47500,code,3.401866368047779 +dflash_gen800k/step_57500,57500,code,3.414100546939387 diff --git a/results/historical-al-curves/v1/gendense_l6/al_curve.csv b/results/historical-al-curves/v1/gendense_l6/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a46f57b9579425b21f4b5adb8c3f0fa8d55a59c5 --- /dev/null +++ b/results/historical-al-curves/v1/gendense_l6/al_curve.csv @@ -0,0 +1,7 @@ +ckpt,step,domain,accept_length +dflash_gen800k/step_8750,8750,code,3.1563344185080005 +dflash_gen800k/step_18750,18750,code,3.278273381294964 +dflash_gen800k/step_28750,28750,code,3.359728673597287 +dflash_gen800k/step_38750,38750,code,3.417429128543573 +dflash_gen800k/step_48750,48750,code,3.4044079193126633 +dflash_gen800k/step_58750,58750,code,3.395782100007452 diff --git a/results/historical-al-curves/v1/gendense_l7/al_curve.csv b/results/historical-al-curves/v1/gendense_l7/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a5619df81d5b8977fae05d236fc08802aa43fced --- /dev/null +++ b/results/historical-al-curves/v1/gendense_l7/al_curve.csv @@ -0,0 +1,7 @@ +ckpt,step,domain,accept_length +dflash_gen800k/step_10000,10000,code,3.1870191635193734 +dflash_gen800k/step_20000,20000,code,3.332212065813528 +dflash_gen800k/step_30000,30000,code,3.3754074074074074 +dflash_gen800k/step_40000,40000,code,3.3776591801942035 +dflash_gen800k/step_50000,50000,code,3.407462798175428 +dflash_gen800k/step_60000,60000,code,3.404662283323371 diff --git a/results/historical-al-curves/v1/genep4_code/al_curve.csv b/results/historical-al-curves/v1/genep4_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..144611431a472b37f5cbe82076d376dd1914c0c5 --- /dev/null +++ b/results/historical-al-curves/v1/genep4_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_gen800k/epoch_4_step_62425,62425,code,3.401358513099948 diff --git a/results/historical-al-curves/v1/genep4_math/al_curve.csv b/results/historical-al-curves/v1/genep4_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..3fb58cc14d4f6d7240bf3c163a1e3bae1aecaeb5 --- /dev/null +++ b/results/historical-al-curves/v1/genep4_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_gen800k/epoch_4_step_62425,62425,math,4.896174863387978 diff --git a/results/historical-al-curves/v1/merge89/al_curve.csv b/results/historical-al-curves/v1/merge89/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..9c2ed878a1096eef00de590a653342211d2310aa --- /dev/null +++ b/results/historical-al-curves/v1/merge89/al_curve.csv @@ -0,0 +1,5 @@ +ckpt,step,domain,accept_length +dflash_overfit_codeMLP_v3/step_15000,15000,code,3.4697327343333586 +dflash_overfit_codeMLP_v3/step_16875,16875,code,3.4550003791037986 +dflash_overfit_codeMLP_v3/step_17500,17500,code,3.4375377187688594 +dflash_overfit_codeMLP_v3/step_18125,18125,code,3.433911077618689 diff --git a/results/historical-al-curves/v1/mv2c89/al_curve.csv b/results/historical-al-curves/v1/mv2c89/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..9b6c9fdfb19b502809cf52d2ee7023e54dfd9582 --- /dev/null +++ b/results/historical-al-curves/v1/mv2c89/al_curve.csv @@ -0,0 +1,11 @@ +ckpt,step,domain,accept_length +dflash_overfit_codeMLP_v2/step_5000,5000,code,3.171933732423778 +dflash_overfit_codeMLP_v2/step_10000,10000,code,3.2386638237384506 +dflash_overfit_codeMLP_v2/step_15000,15000,code,3.3046631372833417 +dflash_overfit_codeMLP_v2/step_20000,20000,code,3.3341625814004536 +dflash_overfit_codeMLP_v2/step_25000,25000,code,3.3779095626389917 +dflash_overfit_codeMLP_v2/step_30000,30000,code,3.3924955330553903 +dflash_overfit_codeMLP_v2/step_35000,35000,code,3.3962882909741374 +dflash_overfit_codeMLP_v2/step_40000,40000,code,3.419993995797058 +dflash_overfit_codeMLP_v2/step_45000,45000,code,3.436241610738255 +dflash_overfit_codeMLP_v2/step_49375,49375,code,3.404662283323371 diff --git a/results/historical-al-curves/v1/mv3x89/al_curve.csv b/results/historical-al-curves/v1/mv3x89/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2865e5a51fa08a57b187f8057411ad41f6dfc3a6 --- /dev/null +++ b/results/historical-al-curves/v1/mv3x89/al_curve.csv @@ -0,0 +1,5 @@ +ckpt,step,domain,accept_length +dflash_overfit_codeMLP_v3/step_6250,6250,code,3.417941794179418 +dflash_overfit_codeMLP_v3/step_8750,8750,code,3.407717618905175 +dflash_overfit_codeMLP_v3/step_11250,11250,code,3.4146122143124766 +dflash_overfit_codeMLP_v3/step_12500,12500,code,3.425650278153661 diff --git a/results/historical-al-curves/v1/mx_code_code/al_curve.csv b/results/historical-al-curves/v1/mx_code_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..b80c369e24e72c0d4d75b4ee818668bcc7edcc18 --- /dev/null +++ b/results/historical-al-curves/v1/mx_code_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/code,0,code,3.687029694959139 diff --git a/results/historical-al-curves/v1/mx_code_creative_writing/al_curve.csv b/results/historical-al-curves/v1/mx_code_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a37fadec26c575bb3b4d8b7c079d0d7ab1849188 --- /dev/null +++ b/results/historical-al-curves/v1/mx_code_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/code,0,creative_writing,2.6923751095530237 diff --git a/results/historical-al-curves/v1/mx_code_factual_qa/al_curve.csv b/results/historical-al-curves/v1/mx_code_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..bf8902e7ff99e34caccb24b360d30b05d1c87217 --- /dev/null +++ b/results/historical-al-curves/v1/mx_code_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/code,0,factual_qa,2.8411461455460976 diff --git a/results/historical-al-curves/v1/mx_code_general/al_curve.csv b/results/historical-al-curves/v1/mx_code_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..653ce534e2df041a687149494e29c85912468ab9 --- /dev/null +++ b/results/historical-al-curves/v1/mx_code_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/code,0,general,3.181580510992276 diff --git a/results/historical-al-curves/v1/mx_code_math/al_curve.csv b/results/historical-al-curves/v1/mx_code_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..adf7bb4e08d7aefc53ab938058950afb1a9b16fa --- /dev/null +++ b/results/historical-al-curves/v1/mx_code_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/code,0,math,4.733227680929741 diff --git a/results/historical-al-curves/v1/mx_creative_writing_code/al_curve.csv b/results/historical-al-curves/v1/mx_creative_writing_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..09e1fe4ef5a564ce2b355fcd87847ba1acb35754 --- /dev/null +++ b/results/historical-al-curves/v1/mx_creative_writing_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/creative_writing,0,code,3.2806335493160548 diff --git a/results/historical-al-curves/v1/mx_creative_writing_creative_writing/al_curve.csv b/results/historical-al-curves/v1/mx_creative_writing_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e3a1212594bf9e6e9b18306fe081ed586a130ee8 --- /dev/null +++ b/results/historical-al-curves/v1/mx_creative_writing_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/creative_writing,0,creative_writing,2.9755908562572646 diff --git a/results/historical-al-curves/v1/mx_creative_writing_factual_qa/al_curve.csv b/results/historical-al-curves/v1/mx_creative_writing_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..abea304e173a3216706bfa245dccfeb6f9447bc7 --- /dev/null +++ b/results/historical-al-curves/v1/mx_creative_writing_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/creative_writing,0,factual_qa,2.9040243902439022 diff --git a/results/historical-al-curves/v1/mx_creative_writing_general/al_curve.csv b/results/historical-al-curves/v1/mx_creative_writing_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e4d346fc9ee6bc0edfbc4308d8ecccbb1bd017aa --- /dev/null +++ b/results/historical-al-curves/v1/mx_creative_writing_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/creative_writing,0,general,3.21130309575234 diff --git a/results/historical-al-curves/v1/mx_creative_writing_math/al_curve.csv b/results/historical-al-curves/v1/mx_creative_writing_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..559f7ba61a7c69e17115569ed014f4097eaf39d5 --- /dev/null +++ b/results/historical-al-curves/v1/mx_creative_writing_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/creative_writing,0,math,4.67396974439228 diff --git a/results/historical-al-curves/v1/mx_factual_qa_code/al_curve.csv b/results/historical-al-curves/v1/mx_factual_qa_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..fbfbb7454670c1b0310f939f8afd6b8c9890f43a --- /dev/null +++ b/results/historical-al-curves/v1/mx_factual_qa_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/factual_qa,0,code,3.282051282051282 diff --git a/results/historical-al-curves/v1/mx_factual_qa_creative_writing/al_curve.csv b/results/historical-al-curves/v1/mx_factual_qa_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e0efc60b526b8d9d465b1f7a6a83f02c7cd2713e --- /dev/null +++ b/results/historical-al-curves/v1/mx_factual_qa_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/factual_qa,0,creative_writing,2.7224388514711095 diff --git a/results/historical-al-curves/v1/mx_factual_qa_factual_qa/al_curve.csv b/results/historical-al-curves/v1/mx_factual_qa_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e0a2f871e2cbad5da48a9e9853ddfc89a3a6bff3 --- /dev/null +++ b/results/historical-al-curves/v1/mx_factual_qa_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/factual_qa,0,factual_qa,3.0694480435345946 diff --git a/results/historical-al-curves/v1/mx_factual_qa_general/al_curve.csv b/results/historical-al-curves/v1/mx_factual_qa_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..01b125c567a3822ddaa553a3383bd131dac3c55d --- /dev/null +++ b/results/historical-al-curves/v1/mx_factual_qa_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/factual_qa,0,general,3.24297480620155 diff --git a/results/historical-al-curves/v1/mx_factual_qa_math/al_curve.csv b/results/historical-al-curves/v1/mx_factual_qa_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..7ba66d84be56b7ec33a4ca2776e8ded519328503 --- /dev/null +++ b/results/historical-al-curves/v1/mx_factual_qa_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/factual_qa,0,math,4.703412073490814 diff --git a/results/historical-al-curves/v1/mx_general_code/al_curve.csv b/results/historical-al-curves/v1/mx_general_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..cccb1d88d034ccaff98c7ee7846bb215fe4e4c96 --- /dev/null +++ b/results/historical-al-curves/v1/mx_general_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/general,0,code,3.35700604096066 diff --git a/results/historical-al-curves/v1/mx_general_creative_writing/al_curve.csv b/results/historical-al-curves/v1/mx_general_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..39a284f7395dabe421ad835615022170eaec7b3e --- /dev/null +++ b/results/historical-al-curves/v1/mx_general_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/general,0,creative_writing,2.8147333699835073 diff --git a/results/historical-al-curves/v1/mx_general_factual_qa/al_curve.csv b/results/historical-al-curves/v1/mx_general_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..cfebfea536d5090cb5755c4b109092131d782f8c --- /dev/null +++ b/results/historical-al-curves/v1/mx_general_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/general,0,factual_qa,2.958063713290194 diff --git a/results/historical-al-curves/v1/mx_general_general/al_curve.csv b/results/historical-al-curves/v1/mx_general_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c5fbcd9f5a9e5ab86eac8cb937200f754739f7b7 --- /dev/null +++ b/results/historical-al-curves/v1/mx_general_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/general,0,general,3.277839437661857 diff --git a/results/historical-al-curves/v1/mx_general_math/al_curve.csv b/results/historical-al-curves/v1/mx_general_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..0d925a3e23dc88793cade9c9a75d34c894cb670b --- /dev/null +++ b/results/historical-al-curves/v1/mx_general_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/general,0,math,4.965364366860626 diff --git a/results/historical-al-curves/v1/mx_math_code/al_curve.csv b/results/historical-al-curves/v1/mx_math_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..5269c41421e83a10f56704e7318c78c23c9c368d --- /dev/null +++ b/results/historical-al-curves/v1/mx_math_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/math,0,code,3.2578823192964896 diff --git a/results/historical-al-curves/v1/mx_math_creative_writing/al_curve.csv b/results/historical-al-curves/v1/mx_math_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..46f308b55b6cb5c8c103ddd9e8c4049ee64b28a8 --- /dev/null +++ b/results/historical-al-curves/v1/mx_math_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/math,0,creative_writing,2.6095820591233436 diff --git a/results/historical-al-curves/v1/mx_math_factual_qa/al_curve.csv b/results/historical-al-curves/v1/mx_math_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..9aaa5815d13655af0583a0ab69fa69892379ca0a --- /dev/null +++ b/results/historical-al-curves/v1/mx_math_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/math,0,factual_qa,2.7249628018770746 diff --git a/results/historical-al-curves/v1/mx_math_general/al_curve.csv b/results/historical-al-curves/v1/mx_math_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..2a907320c4f222a4febb368dbc164fd75c90fb1c --- /dev/null +++ b/results/historical-al-curves/v1/mx_math_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/math,0,general,3.0759333716255024 diff --git a/results/historical-al-curves/v1/mx_math_math/al_curve.csv b/results/historical-al-curves/v1/mx_math_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..a23dfaa5d5007c5ca2e97e0d62d73b1676f35498 --- /dev/null +++ b/results/historical-al-curves/v1/mx_math_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +b5_experts89/math,0,math,5.47677261613692 diff --git a/results/historical-al-curves/v1/rc89/al_curve.csv b/results/historical-al-curves/v1/rc89/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..29eca2d07f53723ba69a7448295b9cf0841e3e3e --- /dev/null +++ b/results/historical-al-curves/v1/rc89/al_curve.csv @@ -0,0 +1,8 @@ +ckpt,step,domain,accept_length +dflash_route_code_d0/step_5000,5000,code,3.2437357630979498 +dflash_route_code_d0/step_10000,10000,code,3.320072859744991 +dflash_route_code_d0/step_15000,15000,code,3.382673892064435 +dflash_route_code_d0/step_20000,20000,code,3.411033760011977 +dflash_route_code_d0/step_25000,25000,code,3.4471593917845524 +dflash_route_code_d0/step_30000,30000,code,3.470789854520527 +dflash_route_code_d0/step_35000,35000,code,3.445074468889393 diff --git a/results/historical-al-curves/v1/route89/al_curve.csv b/results/historical-al-curves/v1/route89/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..df81e30a4c0a02342ea857308248073fe38ab651 --- /dev/null +++ b/results/historical-al-curves/v1/route89/al_curve.csv @@ -0,0 +1,5 @@ +ckpt,step,domain,accept_length +dflash_route_code_d0_wr/step_18750,18750,code,3.5073891625615765 +dflash_route_code_d0_wr/step_21250,21250,code,3.519036219013051 +dflash_route_code_d0_wr/step_21875,21875,code,3.496623695518723 +dflash_route_code_d0_wr/step_23750,23750,code,3.523117365084274 diff --git a/results/historical-al-curves/v1/rw89x/al_curve.csv b/results/historical-al-curves/v1/rw89x/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..ad1d4e8ff08aa020a0625e3fb81f51b42aabb40b --- /dev/null +++ b/results/historical-al-curves/v1/rw89x/al_curve.csv @@ -0,0 +1,4 @@ +ckpt,step,domain,accept_length +dflash_route_code_d0_wr/step_9375,9375,code,3.4792700618462242 +dflash_route_code_d0_wr/step_14375,14375,code,3.489661510185327 +dflash_route_code_d0_wr/step_16875,16875,code,3.495282657053003 diff --git a/results/historical-al-curves/v1/rwr2_0_code/al_curve.csv b/results/historical-al-curves/v1/rwr2_0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6e875707bad3429d2a4cb3183277b9acebbf0977 --- /dev/null +++ b/results/historical-al-curves/v1/rwr2_0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_route_code_wr2/epoch_0_step_6242,6242,code,3.48699112335476 diff --git a/results/historical-al-curves/v1/rwr2_1_code/al_curve.csv b/results/historical-al-curves/v1/rwr2_1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..82dfbe7b787de392fb3455cc761a2786b7440683 --- /dev/null +++ b/results/historical-al-curves/v1/rwr2_1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_route_code_wr2/epoch_1_step_12484,12484,code,3.516592066676956 diff --git a/results/historical-al-curves/v1/rwr2_2_code/al_curve.csv b/results/historical-al-curves/v1/rwr2_2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..fc932181eee183775e0ff6c13539a5c356da4053 --- /dev/null +++ b/results/historical-al-curves/v1/rwr2_2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_route_code_wr2/epoch_2_step_18726,18726,code,3.537064348366064 diff --git a/results/historical-al-curves/v1/scnAg0_code/al_curve.csv b/results/historical-al-curves/v1/scnAg0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..1248d254fdca7a8a43806c81fa233388bd18aa80 --- /dev/null +++ b/results/historical-al-curves/v1/scnAg0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_gen/epoch_0_step_3125,3125,code,3.371411660254513 diff --git a/results/historical-al-curves/v1/scnAg0_math/al_curve.csv b/results/historical-al-curves/v1/scnAg0_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..eea4cd3ff94e280b1cea9a8766a0a672474f62ea --- /dev/null +++ b/results/historical-al-curves/v1/scnAg0_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_gen/epoch_0_step_3125,3125,math,4.951644100580271 diff --git a/results/historical-al-curves/v1/scnAg1_code/al_curve.csv b/results/historical-al-curves/v1/scnAg1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..5e64d08a266d977cc4a194ff807a7e7b0954de38 --- /dev/null +++ b/results/historical-al-curves/v1/scnAg1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_gen/epoch_1_step_6250,6250,code,3.3540409244810836 diff --git a/results/historical-al-curves/v1/scnAg1_math/al_curve.csv b/results/historical-al-curves/v1/scnAg1_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..66ed17d8cb03c370ab53e3cb037b6c80c447fa99 --- /dev/null +++ b/results/historical-al-curves/v1/scnAg1_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_gen/epoch_1_step_6250,6250,math,5.0 diff --git a/results/historical-al-curves/v1/scnAg2_code/al_curve.csv b/results/historical-al-curves/v1/scnAg2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..bc5026e3dd664146df00337b922f3a06e21b3098 --- /dev/null +++ b/results/historical-al-curves/v1/scnAg2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_gen/epoch_2_step_9375,9375,code,3.35873811454264 diff --git a/results/historical-al-curves/v1/scnAg2_math/al_curve.csv b/results/historical-al-curves/v1/scnAg2_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..ae2f0951e681007bca6ee50a4c62d3096edec936 --- /dev/null +++ b/results/historical-al-curves/v1/scnAg2_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_gen/epoch_2_step_9375,9375,math,5.039370078740157 diff --git a/results/historical-al-curves/v1/scnAg3_code/al_curve.csv b/results/historical-al-curves/v1/scnAg3_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..9012fdde8f3561c24aba4fca2c674a4f0da1c9c3 --- /dev/null +++ b/results/historical-al-curves/v1/scnAg3_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_gen/epoch_3_step_12500,12500,code,3.387703516467177 diff --git a/results/historical-al-curves/v1/scnAg3_math/al_curve.csv b/results/historical-al-curves/v1/scnAg3_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..68022a4a31844329b2717af7c14449c2018158ab --- /dev/null +++ b/results/historical-al-curves/v1/scnAg3_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_gen/epoch_3_step_12500,12500,math,5.1097804391217565 diff --git a/results/historical-al-curves/v1/scnAm0_math/al_curve.csv b/results/historical-al-curves/v1/scnAm0_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..cd99ef1dc0b678b77f56ac3c3877491bdcfc913f --- /dev/null +++ b/results/historical-al-curves/v1/scnAm0_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_mos/epoch_0_step_3125,3125,math,5.466748017083588 diff --git a/results/historical-al-curves/v1/scnAm1_math/al_curve.csv b/results/historical-al-curves/v1/scnAm1_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c301e2714622b22a81fba2f701a00cc05e876fa9 --- /dev/null +++ b/results/historical-al-curves/v1/scnAm1_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_mos/epoch_1_step_6250,6250,math,5.433596118859915 diff --git a/results/historical-al-curves/v1/scnAm2_math/al_curve.csv b/results/historical-al-curves/v1/scnAm2_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..565245e8e69f485f355c55ce0cfc248e4c336916 --- /dev/null +++ b/results/historical-al-curves/v1/scnAm2_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_mos/epoch_2_step_9375,9375,math,5.507068223724646 diff --git a/results/historical-al-curves/v1/scnAm3_math/al_curve.csv b/results/historical-al-curves/v1/scnAm3_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..7d71f7279ffd4b12fd52c363a587ef4843841b9e --- /dev/null +++ b/results/historical-al-curves/v1/scnAm3_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_mos/epoch_3_step_12500,12500,math,5.441846340722745 diff --git a/results/historical-al-curves/v1/scnAm_code/al_curve.csv b/results/historical-al-curves/v1/scnAm_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c2bb0e141be6b54f20cb6472594c7e2dd31feab9 --- /dev/null +++ b/results/historical-al-curves/v1/scnAm_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_scnA_mos/epoch_3_step_12500,12500,code,3.228567379906476 diff --git a/results/historical-al-curves/v1/sd0_code/al_curve.csv b/results/historical-al-curves/v1/sd0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6d91d526a7a6c34fce7a4dd3626f0bb9dc7f09ab --- /dev/null +++ b/results/historical-al-curves/v1/sd0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_0_step_6242,6242,code,3.185682326621924 diff --git a/results/historical-al-curves/v1/sd1_code/al_curve.csv b/results/historical-al-curves/v1/sd1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..06ef2d0c119efa63e9332c34145281a67aaead81 --- /dev/null +++ b/results/historical-al-curves/v1/sd1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_1_step_12484,12484,code,3.294866232827187 diff --git a/results/historical-al-curves/v1/sd2_code/al_curve.csv b/results/historical-al-curves/v1/sd2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..9f03433f549e297bb828695f0f1462222dbf2ba3 --- /dev/null +++ b/results/historical-al-curves/v1/sd2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_2_step_18726,18726,code,3.371910611217996 diff --git a/results/historical-al-curves/v1/sd3_code/al_curve.csv b/results/historical-al-curves/v1/sd3_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..8b46b0da3cb2165d401e1316101ed67524355297 --- /dev/null +++ b/results/historical-al-curves/v1/sd3_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_3_step_24968,24968,code,3.420507431316619 diff --git a/results/historical-al-curves/v1/sd4_code/al_curve.csv b/results/historical-al-curves/v1/sd4_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..ba55d0d1f7058e1a075dec75638765b62f3e42e2 --- /dev/null +++ b/results/historical-al-curves/v1/sd4_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_4_step_31210,31210,code,3.44846375056758 diff --git a/results/historical-al-curves/v1/sd5_code/al_curve.csv b/results/historical-al-curves/v1/sd5_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..7212bbd7e76f2691a45b3c899a772a8898f31636 --- /dev/null +++ b/results/historical-al-curves/v1/sd5_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_5_step_37452,37452,code,3.514422335338578 diff --git a/results/historical-al-curves/v1/sd5pk_creative_writing/al_curve.csv b/results/historical-al-curves/v1/sd5pk_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..49da5da23f523f3bca51632a72aa7865176daf02 --- /dev/null +++ b/results/historical-al-curves/v1/sd5pk_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_5_step_37452,37452,creative_writing,2.4396442185514613 diff --git a/results/historical-al-curves/v1/sd5pk_factual_qa/al_curve.csv b/results/historical-al-curves/v1/sd5pk_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..58e8d5c0382c6bb37752368d4935b4bdb2c3c6b1 --- /dev/null +++ b/results/historical-al-curves/v1/sd5pk_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_5_step_37452,37452,factual_qa,2.630009937065253 diff --git a/results/historical-al-curves/v1/sd5pk_general/al_curve.csv b/results/historical-al-curves/v1/sd5pk_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..48317d2ed6dcd09c90cee2e9e95340473f2b0cf1 --- /dev/null +++ b/results/historical-al-curves/v1/sd5pk_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_5_step_37452,37452,general,2.886995902523183 diff --git a/results/historical-al-curves/v1/sd5pk_math/al_curve.csv b/results/historical-al-curves/v1/sd5pk_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..d2ca235e6519c552ed73101f491c495d0f9b066a --- /dev/null +++ b/results/historical-al-curves/v1/sd5pk_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_5_step_37452,37452,math,4.2274121255012975 diff --git a/results/historical-al-curves/v1/sd6_code/al_curve.csv b/results/historical-al-curves/v1/sd6_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..c48d23e4ec871fd6ad836ec1705b10a3733d0266 --- /dev/null +++ b/results/historical-al-curves/v1/sd6_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_6_step_43694,43694,code,3.488325805710786 diff --git a/results/historical-al-curves/v1/sd7_code/al_curve.csv b/results/historical-al-curves/v1/sd7_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..5263d3a7144c4f8b5811b4a2623d11c48c6c4271 --- /dev/null +++ b/results/historical-al-curves/v1/sd7_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sd0_code/epoch_7_step_49936,49936,code,3.4872579781128032 diff --git a/results/historical-al-curves/v1/sg0_code/al_curve.csv b/results/historical-al-curves/v1/sg0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..5b88bb06fd6ccd0e3d9f8b6c4ae9135090d8e5d4 --- /dev/null +++ b/results/historical-al-curves/v1/sg0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_0_step_6242,6242,code,3.4112891151369964 diff --git a/results/historical-al-curves/v1/sg1_code/al_curve.csv b/results/historical-al-curves/v1/sg1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..9494a593f938384506746363242a261a971f3899 --- /dev/null +++ b/results/historical-al-curves/v1/sg1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_1_step_12484,12484,code,3.490463423975488 diff --git a/results/historical-al-curves/v1/sg2_code/al_curve.csv b/results/historical-al-curves/v1/sg2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..f065d4dddc6f2ace9c5f625b69356b2c78f35917 --- /dev/null +++ b/results/historical-al-curves/v1/sg2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_2_step_18726,18726,code,3.519036219013051 diff --git a/results/historical-al-curves/v1/sg3_code/al_curve.csv b/results/historical-al-curves/v1/sg3_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..21289d79caf5693318c0cdb305ca82df199f0ccf --- /dev/null +++ b/results/historical-al-curves/v1/sg3_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_3_step_24968,24968,code,3.5866194411648955 diff --git a/results/historical-al-curves/v1/sg4_code/al_curve.csv b/results/historical-al-curves/v1/sg4_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e47a59c04106b394fb142adc6104c288851ac213 --- /dev/null +++ b/results/historical-al-curves/v1/sg4_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_4_step_31210,31210,code,3.56306200641176 diff --git a/results/historical-al-curves/v1/sg5_code/al_curve.csv b/results/historical-al-curves/v1/sg5_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..ae2a8a01c24fa27b2ac8c88aa9bfed4ce1ca812e --- /dev/null +++ b/results/historical-al-curves/v1/sg5_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_5_step_37452,37452,code,3.6073464217859406 diff --git a/results/historical-al-curves/v1/sg6_code/al_curve.csv b/results/historical-al-curves/v1/sg6_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..12a23bda48c6a118e6b529c81f7132aa5ff678d2 --- /dev/null +++ b/results/historical-al-curves/v1/sg6_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_6_step_43694,43694,code,3.6182309036048914 diff --git a/results/historical-al-curves/v1/sg7_code/al_curve.csv b/results/historical-al-curves/v1/sg7_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..afe707c770ff346fd7bbbbc1453c42590253797e --- /dev/null +++ b/results/historical-al-curves/v1/sg7_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_7_step_49936,49936,code,3.5807009272355805 diff --git a/results/historical-al-curves/v1/sgpk_creative_writing/al_curve.csv b/results/historical-al-curves/v1/sgpk_creative_writing/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..e4b244aeb8253cd17b98ebec29fdd9c027c2ceed --- /dev/null +++ b/results/historical-al-curves/v1/sgpk_creative_writing/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_7_step_49936,49936,creative_writing,2.6778242677824267 diff --git a/results/historical-al-curves/v1/sgpk_factual_qa/al_curve.csv b/results/historical-al-curves/v1/sgpk_factual_qa/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..7616f90f0083af038e7f1d1cdb2396981b07b489 --- /dev/null +++ b/results/historical-al-curves/v1/sgpk_factual_qa/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_7_step_49936,49936,factual_qa,2.8229438255510786 diff --git a/results/historical-al-curves/v1/sgpk_general/al_curve.csv b/results/historical-al-curves/v1/sgpk_general/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..baffa65fed99e8acbea8ab9fcc27ada59abbebb2 --- /dev/null +++ b/results/historical-al-curves/v1/sgpk_general/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_7_step_49936,49936,general,3.132669983416252 diff --git a/results/historical-al-curves/v1/sgpk_math/al_curve.csv b/results/historical-al-curves/v1/sgpk_math/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..3718b63ac698f20233902cfed1c3b0a8ed44e59d --- /dev/null +++ b/results/historical-al-curves/v1/sgpk_math/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_sgen_code/epoch_7_step_49936,49936,math,4.702177906061401 diff --git a/results/historical-al-curves/v1/wr2be0_code/al_curve.csv b/results/historical-al-curves/v1/wr2be0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..1211ddddb03f9676a5120c2488cb7dbbf7e1039a --- /dev/null +++ b/results/historical-al-curves/v1/wr2be0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_routewarm_code_wr2b/epoch_0_step_12484,12484,code,3.5619479402798406 diff --git a/results/historical-al-curves/v1/wr2be1_code/al_curve.csv b/results/historical-al-curves/v1/wr2be1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..9e8b58ae46c9023cc337b9ff8da9b202f1859265 --- /dev/null +++ b/results/historical-al-curves/v1/wr2be1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_routewarm_code_wr2b/epoch_1_step_24968,24968,code,3.5555555555555554 diff --git a/results/historical-al-curves/v1/wr2be2_code/al_curve.csv b/results/historical-al-curves/v1/wr2be2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..bffa6b3a00b06fe556fcb588073db160ab06f082 --- /dev/null +++ b/results/historical-al-curves/v1/wr2be2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_routewarm_code_wr2b/epoch_2_step_37452,37452,code,3.579294635142565 diff --git a/results/historical-al-curves/v1/wr3e0_code/al_curve.csv b/results/historical-al-curves/v1/wr3e0_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..1336e3f8cdd944ca5a0e220c6fd0a1404d3886ff --- /dev/null +++ b/results/historical-al-curves/v1/wr3e0_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_route_code_wr3/epoch_0_step_12484,12484,code,3.5179495097660776 diff --git a/results/historical-al-curves/v1/wr3e1_code/al_curve.csv b/results/historical-al-curves/v1/wr3e1_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..6ce210f6d7208ee3438d7bf4a0b9818739270fcc --- /dev/null +++ b/results/historical-al-curves/v1/wr3e1_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_route_code_wr3/epoch_1_step_24968,24968,code,3.510090895085503 diff --git a/results/historical-al-curves/v1/wr3e2_code/al_curve.csv b/results/historical-al-curves/v1/wr3e2_code/al_curve.csv new file mode 100644 index 0000000000000000000000000000000000000000..0b3458ece600d13deea9fb24f539099ca951e893 --- /dev/null +++ b/results/historical-al-curves/v1/wr3e2_code/al_curve.csv @@ -0,0 +1,2 @@ +ckpt,step,domain,accept_length +dflash_route_code_wr3/epoch_2_step_37452,37452,code,3.5439415150101103 diff --git a/results/main-figure-r1/v1/frozen-aggregate-cells.json b/results/main-figure-r1/v1/frozen-aggregate-cells.json new file mode 100644 index 0000000000000000000000000000000000000000..f42328687e13131cb6f96773eb2fe1b71986cf5c --- /dev/null +++ b/results/main-figure-r1/v1/frozen-aggregate-cells.json @@ -0,0 +1,90 @@ +{ + "analysis": "r1_mainfig_cells_summary", + "cells_passed": 52, + "cells_total": 52, + "created_utc": "2026-07-21T07:05:33.502902+00:00", + "panel_b_matrix_dflash_init": { + "code": { + "code": 3.655089436111334, + "creative_writing": 2.535490260812149, + "factual_qa": 2.765586903517938, + "general": 3.0114447592067988, + "math": 4.39000489955904 + }, + "creative_writing": { + "code": 2.985911801323635, + "creative_writing": 2.927945101029356, + "factual_qa": 2.7863077823288473, + "general": 3.0752314814814814, + "math": 4.126180059866452 + }, + "factual_qa": { + "code": 3.064218949633515, + "creative_writing": 2.6560608680615596, + "factual_qa": 2.9658779576587797, + "general": 3.0955985095482066, + "math": 4.351627003399709 + }, + "general": { + "code": 3.174585481398913, + "creative_writing": 2.688604936110625, + "factual_qa": 2.9006574141709276, + "general": 3.1881461889877376, + "math": 4.743250397035468 + }, + "math": { + "code": 3.00639968331464, + "creative_writing": 2.3813953488372093, + "factual_qa": 2.5630650021523893, + "general": 2.8860896948637205, + "math": 5.4057315233785825 + } + }, + "panel_c_matrix_warm_start": { + "code": { + "code": 3.683155512447462, + "creative_writing": 2.668056279312142, + "factual_qa": 2.9102579777478907, + "general": 3.2091463414634145, + "math": 4.735729386892178 + }, + "creative_writing": { + "code": 3.2882089767643237, + "creative_writing": 2.966396292004635, + "factual_qa": 2.9351338019484525, + "general": 3.2797384007897334, + "math": 4.618556701030927 + }, + "factual_qa": { + "code": 3.331724793448856, + "creative_writing": 2.7477638640429336, + "factual_qa": 3.069116698903933, + "general": 3.246088019559902, + "math": 4.713308784850079 + }, + "general": { + "code": 3.335626967279116, + "creative_writing": 2.7645788336933044, + "factual_qa": 2.9983623078861172, + "general": 3.333794839521712, + "math": 5.021014289717008 + }, + "math": { + "code": 3.237053349435249, + "creative_writing": 2.6055979643765905, + "factual_qa": 2.826226012793177, + "general": 3.089057928613224, + "math": 5.560037232392181 + } + }, + "panel_c_note": "diagonal = reused locked B2 domain-assigned cells; off-diagonal = this round", + "panel_d_generalist": { + "code": 3.4346875706640536, + "creative_writing": 2.7321237993596585, + "factual_qa": 2.915187376725838, + "general": 3.158226343319068, + "math": 4.920373421197144 + }, + "passed": true, + "server_random_seed": 20260719 +} diff --git a/verification/v1/bootstrap_b1_exact_3p2m_routed.py b/verification/v1/bootstrap_b1_exact_3p2m_routed.py new file mode 100644 index 0000000000000000000000000000000000000000..694f3c5a11e2a088afe13b10a86608e1299595b1 --- /dev/null +++ b/verification/v1/bootstrap_b1_exact_3p2m_routed.py @@ -0,0 +1,419 @@ +#!/usr/bin/env python3 +"""Validate and bootstrap exact-3.2M routed MoS against B1 references. + +Only the five router-selected MoS buckets are newly generated. The exact B1 +domain-assigned MoS and generalist sidecars are reused after their file hashes, +prompt digests, counters, checkpoints, and fixed SGLang seed are revalidated. +All comparisons are paired by the exact loaded-question digest and regrouped by the +offline ReasonMix assignment before 50k-stratified bootstrap resampling. +""" + +from __future__ import annotations + +import argparse +import json +import math +import os +from datetime import datetime, timezone +from pathlib import Path +from typing import Any + +import numpy as np + +from bootstrap_b2_epoch5_router import ( + DOMAINS, + EXPERT_DIR, + aggregate_al, + condition_summary, + digest_sequence, + expected_prompt_digest, + read_jsonl, + resolve_single, +) +from bootstrap_paired_al import ( + _file_sha256, + _ratio, + _summarize_pair, +) + + +LOCKED_KEYS = [ + "target_model_path", + "draft_model_path", + "source_checkpoint", + "algorithm", + "batch_size", + "steps", + "topk", + "num_draft_tokens", + "attention_backend", + "dtype", + "trust_remote_code", + "temperature", + "max_new_tokens", + "prompt_source", + "bench_chat_template", + "bench_preformat_tokenizer", + "thinking", + "server_random_seed_requested", + "server_random_seed_effective", +] + + +def validate_metric_rows( + source_rows: list[dict[str, Any]], + metric_rows: list[dict[str, Any]], + source_path: Path, + sidecar_path: Path, + expected: dict[str, Any], +) -> tuple[dict[str, tuple[float, float]], list[str], list[str]]: + if len(source_rows) != len(metric_rows): + raise ValueError( + f"{sidecar_path}: {len(metric_rows)} metrics != {len(source_rows)} inputs" + ) + joined: dict[str, tuple[float, float]] = {} + question_digests: list[str] = [] + row_digests: list[str] = [] + for row_idx, (source, metric) in enumerate( + zip(source_rows, metric_rows, strict=True) + ): + label = f"{sidecar_path}:{row_idx + 1}" + if int(metric.get("run_idx", -1)) != 0: + raise ValueError(f"{label}: run_idx must be 0") + if int(metric.get("prompt_idx", -1)) != row_idx: + raise ValueError(f"{label}: prompt_idx is not contiguous input order") + for key in LOCKED_KEYS: + if metric.get(key) != expected[key]: + raise ValueError( + f"{label}: {key}={metric.get(key)!r}, expected {expected[key]!r}" + ) + prompt_digest = expected_prompt_digest(source, source_path, row_idx + 1) + if metric.get("prompt_digest") != prompt_digest: + raise ValueError(f"{label}: prompt digest differs from loaded question") + try: + completion = float(metric["completion_tokens"]) + verify = float(metric["spec_verify_ct"]) + except (KeyError, TypeError, ValueError) as error: + raise ValueError(f"{label}: invalid AL counters") from error + if ( + not math.isfinite(completion) + or not math.isfinite(verify) + or completion < 0 + or verify <= 0 + ): + raise ValueError(f"{label}: invalid counters {completion=}, {verify=}") + digest = prompt_digest + if digest in joined: + raise ValueError(f"duplicate canonical source row: {digest}") + joined[digest] = (completion, verify) + question_digests.append(prompt_digest) + row_digests.append(digest) + return joined, question_digests, row_digests + + +def load_b1_condition( + b1: dict[str, Any], + side: str, + target: Path, + expected_checkpoint: Path, + server_seed: int, +) -> tuple[dict[str, tuple[float, float]], dict[str, int], dict[str, Any]]: + joined: dict[str, tuple[float, float]] = {} + truth: dict[str, int] = {} + provenance: dict[str, Any] = {} + recorded = b1["validated_sidecar_provenance"][side] + for domain_id, domain in enumerate(DOMAINS): + item = recorded[domain] + source_path = Path(item["prompt_source"]).resolve() + sidecar_path = Path(item["sidecar"]).resolve() + result_path = Path(item["aggregate_result"]).resolve() + for path, sha_key in [ + (source_path, "prompt_source_sha256"), + (sidecar_path, "sidecar_sha256"), + (result_path, "aggregate_result_sha256"), + ]: + observed = _file_sha256(path) + if observed != item[sha_key]: + raise ValueError(f"{path}: sha256 {observed} != B1 {item[sha_key]}") + expected = { + "target_model_path": str(target), + "draft_model_path": item["draft_model_path"], + "source_checkpoint": str(expected_checkpoint), + "algorithm": "DFLASH", + "batch_size": 8, + "steps": 1, + "topk": 1, + "num_draft_tokens": 16, + "attention_backend": "fa3", + "dtype": "auto", + "trust_remote_code": True, + "temperature": 0, + "max_new_tokens": 512, + "prompt_source": str(source_path), + "bench_chat_template": None, + "bench_preformat_tokenizer": str(target), + "thinking": True, + "server_random_seed_requested": server_seed, + "server_random_seed_effective": server_seed, + } + source_rows = read_jsonl(source_path) + metric_rows = read_jsonl(sidecar_path) + cell, question_digests, row_digests = validate_metric_rows( + source_rows, metric_rows, source_path, sidecar_path, expected + ) + if set(joined).intersection(cell): + raise ValueError(f"{side}: duplicate rows across true-domain files") + joined.update(cell) + for digest in cell: + truth[digest] = domain_id + pooled = _ratio(np.asarray(list(cell.values()), dtype=np.float64)) + aggregate = aggregate_al(result_path) + if not math.isclose(pooled, aggregate, rel_tol=0.0, abs_tol=1e-10): + raise ValueError(f"{result_path}: sidecar AL {pooled} != aggregate {aggregate}") + provenance[domain] = { + "prompt_source": str(source_path), + "prompt_source_sha256": item["prompt_source_sha256"], + "sidecar": str(sidecar_path), + "sidecar_sha256": item["sidecar_sha256"], + "aggregate_result": str(result_path), + "aggregate_result_sha256": item["aggregate_result_sha256"], + "n_prompts": len(metric_rows), + "pooled_al": pooled, + "prompt_digest_sequence_sha256": digest_sequence(question_digests), + "canonical_row_digest_sequence_sha256": digest_sequence(row_digests), + "validated_harness": expected, + } + return joined, truth, provenance + + +def load_routed( + groups_root: Path, + bench_root: Path, + target: Path, + expert_bank: Path, + exports_root: Path, + server_seed: int, +) -> tuple[dict[str, tuple[float, float]], dict[str, Any]]: + joined: dict[str, tuple[float, float]] = {} + provenance: dict[str, Any] = {} + for domain_id, domain in enumerate(DOMAINS): + group_path = (groups_root / "routed_groups" / f"{domain_id}.jsonl").resolve() + cell_dir = bench_root / "routed" / str(domain_id) + sidecar_path = resolve_single(cell_dir, "*perprompt*.jsonl").resolve() + result_path = resolve_single(cell_dir, "*_results_*.jsonl").resolve() + export_dir = (exports_root / EXPERT_DIR[domain]).resolve() + expected = { + "target_model_path": str(target), + "draft_model_path": str(export_dir), + "source_checkpoint": str(expert_bank), + "algorithm": "DFLASH", + "batch_size": 8, + "steps": 1, + "topk": 1, + "num_draft_tokens": 16, + "attention_backend": "fa3", + "dtype": "auto", + "trust_remote_code": True, + "temperature": 0, + "max_new_tokens": 512, + "prompt_source": str(group_path), + "bench_chat_template": None, + "bench_preformat_tokenizer": str(target), + "thinking": True, + "server_random_seed_requested": server_seed, + "server_random_seed_effective": server_seed, + } + source_rows = read_jsonl(group_path) + metric_rows = read_jsonl(sidecar_path) + cell, question_digests, row_digests = validate_metric_rows( + source_rows, metric_rows, group_path, sidecar_path, expected + ) + if set(joined).intersection(cell): + raise ValueError("routed buckets contain duplicate source rows") + joined.update(cell) + pooled = _ratio(np.asarray(list(cell.values()), dtype=np.float64)) + aggregate = aggregate_al(result_path) + if not math.isclose(pooled, aggregate, rel_tol=0.0, abs_tol=1e-10): + raise ValueError(f"{result_path}: sidecar AL {pooled} != aggregate {aggregate}") + provenance[domain] = { + "router_selected_domain_id": domain_id, + "group": str(group_path), + "group_sha256": _file_sha256(group_path), + "sidecar": str(sidecar_path), + "sidecar_sha256": _file_sha256(sidecar_path), + "aggregate_result": str(result_path), + "aggregate_result_sha256": _file_sha256(result_path), + "n_prompts": len(metric_rows), + "pooled_al_within_router_bucket": pooled, + "prompt_digest_sequence_sha256": digest_sequence(question_digests), + "canonical_row_digest_sequence_sha256": digest_sequence(row_digests), + "validated_harness": expected, + } + return joined, provenance + + +def arrays_by_truth( + values: dict[str, tuple[float, float]], truth: dict[str, int] +) -> list[np.ndarray]: + if set(values) != set(truth): + raise ValueError("condition and offline truth populations differ") + arrays: list[np.ndarray] = [] + for domain_id in range(5): + keys = sorted(key for key, value in truth.items() if value == domain_id) + arrays.append(np.asarray([values[key] for key in keys], dtype=np.float64)) + return arrays + + +def validate_b1_points( + b1: dict[str, Any], domain_assigned: list[np.ndarray], generalist: list[np.ndarray] +) -> None: + expected = b1["results"] + for domain, left, right in zip( + DOMAINS, domain_assigned, generalist, strict=True + ): + item = expected["per_domain"][domain] + for observed, key in [(_ratio(left), "left_al"), (_ratio(right), "right_al")]: + if not math.isclose(observed, float(item[key]), rel_tol=0.0, abs_tol=1e-12): + raise ValueError(f"B1 {domain} {key}: {observed} != {item[key]}") + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--groups-root", required=True, type=Path) + parser.add_argument("--bench-root", required=True, type=Path) + parser.add_argument("--b1-bootstrap", required=True, type=Path) + parser.add_argument("--identity-artifact", required=True, type=Path) + parser.add_argument("--target", required=True, type=Path) + parser.add_argument("--expert-bank", required=True, type=Path) + parser.add_argument("--generalist-checkpoint", required=True, type=Path) + parser.add_argument("--selected-router", required=True, type=Path) + parser.add_argument("--exports-root", required=True, type=Path) + parser.add_argument("--server-random-seed", required=True, type=int) + parser.add_argument("--replicates", type=int, default=50_000) + parser.add_argument("--seed", type=int, default=20_260_719) + parser.add_argument("--chunk-size", type=int, default=500) + parser.add_argument("--output", required=True, type=Path) + args = parser.parse_args() + + output = args.output.resolve() + if output.exists(): + raise FileExistsError(f"refusing to overwrite {output}") + target = args.target.resolve() + expert_bank = args.expert_bank.resolve() + generalist_checkpoint = args.generalist_checkpoint.resolve() + exports_root = args.exports_root.resolve() + bench_root = args.bench_root.resolve() + groups_root = args.groups_root.resolve() + identity_path = args.identity_artifact.resolve() + identity = json.loads(identity_path.read_text()) + if Path(identity["served_expert_bank"]["path"]).resolve() != expert_bank: + raise ValueError("identity artifact records a different served expert bank") + if Path(identity["selected_mean_only_sidecar"]["path"]).resolve() != args.selected_router.resolve(): + raise ValueError("identity artifact records a different selected router") + + b1_path = args.b1_bootstrap.resolve() + b1 = json.loads(b1_path.read_text()) + domain_assigned_values, truth, b1_mos_provenance = load_b1_condition( + b1, "left", target, expert_bank, args.server_random_seed + ) + generalist_values, generalist_truth, b1_gen_provenance = load_b1_condition( + b1, "right", target, generalist_checkpoint, args.server_random_seed + ) + if truth != generalist_truth: + raise ValueError("B1 MoS and generalist offline domain assignments differ") + routed_values, routed_provenance = load_routed( + groups_root, + bench_root, + target, + expert_bank, + exports_root, + args.server_random_seed, + ) + if set(routed_values) != set(truth): + raise ValueError("routed and B1 prompt populations differ") + population_digest = digest_sequence(sorted(truth)) + expected_digest = identity["b1_evaluation_population"][ + "canonical_loaded_question_digest_set_sha256" + ] + if population_digest != expected_digest: + raise ValueError(f"population digest {population_digest} != identity {expected_digest}") + + arrays = { + "routed_mos": arrays_by_truth(routed_values, truth), + "domain_assigned_mos": arrays_by_truth(domain_assigned_values, truth), + "generalist": arrays_by_truth(generalist_values, truth), + } + validate_b1_points(b1, arrays["domain_assigned_mos"], arrays["generalist"]) + + comparisons = {} + for left, right in [ + ("routed_mos", "generalist"), + ("routed_mos", "domain_assigned_mos"), + ("domain_assigned_mos", "generalist"), + ]: + comparisons[f"{left}_vs_{right}"] = { + "left": left, + "right": right, + **_summarize_pair( + arrays[left], + arrays[right], + DOMAINS, + np.random.default_rng(args.seed), + args.replicates, + args.chunk_size, + ), + } + + result = { + "analysis": "b1_exact_3p2m_unified_router_paired_acceptance_length", + "created_utc": datetime.now(timezone.utc).isoformat(), + "numpy_version": np.__version__, + "served_expert_bank_checkpoint": str(expert_bank), + "generalist_checkpoint": str(generalist_checkpoint), + "selected_router_checkpoint": str(args.selected_router.resolve()), + "standalone_exports_root": str(exports_root), + "server_random_seed": args.server_random_seed, + "identity_artifact": str(identity_path), + "identity_artifact_sha256": _file_sha256(identity_path), + "reused_b1_bootstrap_artifact": str(b1_path), + "reused_b1_bootstrap_artifact_sha256": _file_sha256(b1_path), + "groups_root": str(groups_root), + "bench_root": str(bench_root), + "prompt_population": { + "n_prompts": len(truth), + "canonical_loaded_question_digest_set_sha256": population_digest, + }, + "assignment_agreement": identity["assignment_set"][ + "selected_router_recomputed" + ], + "bootstrap": { + "replicates": args.replicates, + "seed": args.seed, + "chunk_size": args.chunk_size, + "interval": "two-sided 95% percentile", + "resampling": "paired by exact benchmark prompt_digest and stratified by offline assignment", + "estimator": "sum(completion_tokens) / sum(spec_verify_ct) in every replicate", + }, + "conditions": { + name: condition_summary(values) for name, values in arrays.items() + }, + "comparisons": comparisons, + "validated_sidecar_provenance": { + "routed_mos": routed_provenance, + "domain_assigned_mos_reused_from_b1": b1_mos_provenance, + "generalist_reused_from_b1": b1_gen_provenance, + }, + "limitations": [ + "Bootstrap intervals quantify evaluation-prompt uncertainty under one fixed serving seed, not training-seed uncertainty.", + "The selected router head was selected and assessed on the same 256 offline assignments.", + ], + } + output.parent.mkdir(parents=True, exist_ok=True) + temp = output.with_suffix(output.suffix + ".tmp") + temp.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + os.replace(temp, output) + print(f"wrote {output}") + + +if __name__ == "__main__": + main() diff --git a/verification/v1/bootstrap_b2_epoch5_router.py b/verification/v1/bootstrap_b2_epoch5_router.py new file mode 100644 index 0000000000000000000000000000000000000000..edfd2a564cdc551346c66802db272897dd6a7807 --- /dev/null +++ b/verification/v1/bootstrap_b2_epoch5_router.py @@ -0,0 +1,390 @@ +#!/usr/bin/env python3 +"""Validate and bootstrap the selected epoch-5 routed-vs-domain-assigned run. + +Rows are joined by a canonical digest of the original source JSONL object even +though routed and domain-assigned cells use different bucket layouts. Each +sidecar's ``prompt_digest`` is also checked against the exact question object +loaded by ``CustomJsonlBenchmarker``. Acceptance length is recomputed as +``sum(completion_tokens) / sum(spec_verify_ct)`` in every bootstrap replicate. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +from datetime import datetime, timezone +from pathlib import Path +from typing import Any + +import numpy as np + +from bootstrap_paired_al import ( + _canonical_digest, + _file_sha256, + _ratio, + _summarize_pair, +) + + +DOMAINS = ["code", "math", "factual_qa", "creative_writing", "general"] +GROUP_DIR = {"routed": "routed_groups", "domain_assigned": "oracle_groups"} +EXPERT_DIR = { + "code": "code", + "math": "math", + "factual_qa": "factual_qa", + "creative_writing": "creative_writing", + "general": "general", +} + + +def read_jsonl(path: Path) -> list[dict[str, Any]]: + rows: list[dict[str, Any]] = [] + with path.open() as handle: + for line_no, line in enumerate(handle, 1): + if not line.strip(): + continue + value = json.loads(line) + if not isinstance(value, dict): + raise ValueError(f"{path}:{line_no}: expected JSON object") + rows.append(value) + return rows + + +def extract_first_user(conversations: list[dict[str, Any]]) -> str | None: + for turn in conversations: + if turn.get("role") == "user": + return turn.get("content", "") + for turn in conversations: + if turn.get("role") != "system": + return turn.get("content", "") + return None + + +def expected_prompt_digest(source: dict[str, Any], path: Path, line_no: int) -> str: + conversations = source.get("conversations", []) + if not isinstance(conversations, list): + raise ValueError(f"{path}:{line_no}: conversations must be a list") + question = extract_first_user(conversations) + if not question: + raise ValueError(f"{path}:{line_no}: benchmark loader would skip this row") + return _canonical_digest({"question": question}) + + +def digest_sequence(values: list[str]) -> str: + return hashlib.sha256(("\n".join(values) + "\n").encode("ascii")).hexdigest() + + +def resolve_single(directory: Path, pattern: str) -> Path: + matches = sorted(directory.glob(pattern)) + if len(matches) != 1: + raise ValueError(f"expected one {pattern} under {directory}, found {len(matches)}") + return matches[0] + + +def aggregate_al(path: Path) -> float: + payload = json.loads(path.read_text()) + if not isinstance(payload, dict): + raise ValueError(f"{path}: expected one JSON result object") + entries = payload.get("customjsonl") + if not isinstance(entries, list) or len(entries) != 1: + raise ValueError(f"{path}: expected one customjsonl result") + entry = entries[0] + locked = (8, 1, 1, 16) + observed = tuple( + int(entry[name]) for name in ["batch_size", "steps", "topk", "num_draft_tokens"] + ) + if observed != locked: + raise ValueError(f"{path}: config {observed} != {locked}") + metrics = entry.get("metrics") + if not isinstance(metrics, list) or len(metrics) != 1: + raise ValueError(f"{path}: expected one metric run") + return float(metrics[0]["accept_length"]) + + +def expected_provenance( + condition: str, + domain_id: int, + group_path: Path, + target: Path, + expert_bank: Path, + exports_root: Path, + server_random_seed: int, +) -> dict[str, Any]: + served_domain = domain_id + export_dir = exports_root / EXPERT_DIR[DOMAINS[served_domain]] + return { + "target_model_path": str(target.resolve()), + "draft_model_path": str(export_dir.resolve()), + "source_checkpoint": str(expert_bank.resolve()), + "algorithm": "DFLASH", + "batch_size": 8, + "steps": 1, + "topk": 1, + "num_draft_tokens": 16, + "attention_backend": "fa3", + "dtype": "bfloat16", + "trust_remote_code": True, + "temperature": 0, + "max_new_tokens": 512, + "prompt_source": str(group_path.resolve()), + "bench_chat_template": None, + "bench_preformat_tokenizer": str(target.resolve()), + "thinking": True, + "server_random_seed_requested": server_random_seed, + "server_random_seed_effective": server_random_seed, + } + + +def load_condition( + condition: str, + groups_root: Path, + bench_root: Path, + target: Path, + expert_bank: Path, + exports_root: Path, + server_random_seed: int, +) -> tuple[dict[str, tuple[float, float]], dict[str, Any]]: + joined: dict[str, tuple[float, float]] = {} + provenance: dict[str, Any] = {} + for domain_id, domain in enumerate(DOMAINS): + group_path = groups_root / GROUP_DIR[condition] / f"{domain_id}.jsonl" + source_rows = read_jsonl(group_path) + cell_dir = bench_root / condition / str(domain_id) + sidecar_path = resolve_single(cell_dir, "*perprompt*.jsonl") + result_path = resolve_single(cell_dir, "*_results_*.jsonl") + metric_rows = read_jsonl(sidecar_path) + if len(source_rows) != len(metric_rows): + raise ValueError( + f"{condition}/{domain_id}: {len(source_rows)} inputs != " + f"{len(metric_rows)} sidecar rows" + ) + expected = expected_provenance( + condition, + domain_id, + group_path, + target, + expert_bank, + exports_root, + server_random_seed, + ) + counts: list[tuple[float, float]] = [] + question_digests: list[str] = [] + row_digests: list[str] = [] + for row_idx, (source, metric) in enumerate( + zip(source_rows, metric_rows, strict=True) + ): + source_label = f"{sidecar_path}:{row_idx + 1}" + if int(metric.get("run_idx", -1)) != 0: + raise ValueError(f"{source_label}: run_idx must be 0") + if int(metric.get("prompt_idx", -1)) != row_idx: + raise ValueError(f"{source_label}: prompt_idx is not contiguous input order") + for key, expected_value in expected.items(): + if metric.get(key) != expected_value: + raise ValueError( + f"{source_label}: {key}={metric.get(key)!r}, " + f"expected {expected_value!r}" + ) + prompt_digest = expected_prompt_digest(source, group_path, row_idx + 1) + if metric.get("prompt_digest") != prompt_digest: + raise ValueError(f"{source_label}: prompt_digest differs from loaded question") + try: + completion = float(metric["completion_tokens"]) + verify = float(metric["spec_verify_ct"]) + except (KeyError, TypeError, ValueError) as error: + raise ValueError(f"{source_label}: invalid AL count fields") from error + if not math.isfinite(completion) or not math.isfinite(verify) or completion < 0 or verify <= 0: + raise ValueError( + f"{source_label}: invalid counts completion={completion}, verify={verify}" + ) + digest = _canonical_digest(source) + if digest in joined: + raise ValueError(f"{condition}: duplicate canonical source-row digest {digest}") + joined[digest] = (completion, verify) + counts.append((completion, verify)) + question_digests.append(prompt_digest) + row_digests.append(digest) + pooled = _ratio(np.asarray(counts, dtype=np.float64)) + recorded = aggregate_al(result_path) + if not math.isclose(pooled, recorded, rel_tol=0.0, abs_tol=1e-10): + raise ValueError( + f"{cell_dir}: sidecar pooled AL {pooled} != aggregate AL {recorded}" + ) + provenance[domain] = { + "group": str(group_path.resolve()), + "group_sha256": _file_sha256(group_path), + "sidecar": str(sidecar_path.resolve()), + "sidecar_sha256": _file_sha256(sidecar_path), + "aggregate_result": str(result_path.resolve()), + "aggregate_result_sha256": _file_sha256(result_path), + "n_prompts": len(metric_rows), + "pooled_al": pooled, + "prompt_digest_sequence_sha256": digest_sequence(question_digests), + "canonical_row_digest_sequence_sha256": digest_sequence(row_digests), + "validated_harness": expected, + } + return joined, provenance + + +def validate_assignment( + groups_root: Path, + confusion_path: Path, + population: set[str], +) -> tuple[dict[str, int], dict[str, int], dict[str, Any]]: + truth: dict[str, int] = {} + prediction: dict[str, int] = {} + for domain_id in range(5): + for condition, target in [("oracle_groups", truth), ("routed_groups", prediction)]: + for row in read_jsonl(groups_root / condition / f"{domain_id}.jsonl"): + digest = _canonical_digest(row) + if digest in target: + raise ValueError(f"duplicate assignment digest in {condition}: {digest}") + target[digest] = domain_id + if set(truth) != population or set(prediction) != population: + raise ValueError("assignment groups do not match evaluated prompt population") + confusion = [[0 for _ in range(5)] for _ in range(5)] + for digest, true_domain in truth.items(): + confusion[true_domain][prediction[digest]] += 1 + artifact = json.loads(confusion_path.read_text()) + if artifact.get("confusion") != confusion: + raise ValueError(f"{confusion_path}: confusion does not match group membership") + correct = sum(confusion[index][index] for index in range(5)) + return truth, prediction, { + "artifact": str(confusion_path.resolve()), + "artifact_sha256": _file_sha256(confusion_path), + "validated_against_group_membership": True, + "overall_assignment_agreement": correct / len(population), + "correct": correct, + "total": len(population), + "confusion": confusion, + "per_domain_recall": [confusion[i][i] / sum(confusion[i]) for i in range(5)], + } + + +def condition_summary(arrays: list[np.ndarray]) -> dict[str, Any]: + per_domain = [_ratio(value) for value in arrays] + return { + "n_prompts": int(sum(value.shape[0] for value in arrays)), + "token_weighted_pooled_al": _ratio(np.concatenate(arrays, axis=0)), + "unweighted_domain_mean_al": float(np.mean(per_domain)), + "per_domain_al": dict(zip(DOMAINS, per_domain, strict=True)), + } + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--groups-root", required=True, type=Path) + parser.add_argument("--bench-root", required=True, type=Path) + parser.add_argument("--confusion-json", required=True, type=Path) + parser.add_argument("--target", required=True, type=Path) + parser.add_argument("--expert-bank", required=True, type=Path) + parser.add_argument("--selected-router", required=True, type=Path) + parser.add_argument("--exports-root", required=True, type=Path) + parser.add_argument("--identity-artifact", required=True, type=Path) + parser.add_argument("--replicates", type=int, default=50_000) + parser.add_argument("--seed", type=int, default=20_260_719) + parser.add_argument("--chunk-size", type=int, default=500) + parser.add_argument("--server-random-seed", required=True, type=int) + parser.add_argument("--output", required=True, type=Path) + args = parser.parse_args() + + output = args.output.resolve() + if output.exists(): + raise FileExistsError(f"refusing to overwrite {output}") + identity = json.loads(args.identity_artifact.read_text()) + if Path(identity["expert_bank"]["path"]).resolve() != args.expert_bank.resolve(): + raise ValueError("identity artifact expert bank differs from requested expert bank") + if Path(identity["selected_router"]["path"]).resolve() != args.selected_router.resolve(): + raise ValueError("identity artifact selected router differs from requested selected router") + + loaded: dict[str, dict[str, tuple[float, float]]] = {} + sidecar_provenance: dict[str, Any] = {} + for condition in ["routed", "domain_assigned"]: + loaded[condition], sidecar_provenance[condition] = load_condition( + condition, + args.groups_root.resolve(), + args.bench_root.resolve(), + args.target.resolve(), + args.expert_bank.resolve(), + args.exports_root.resolve(), + args.server_random_seed, + ) + if set(loaded["routed"]) != set(loaded["domain_assigned"]): + raise ValueError("routed and domain-assigned prompt populations differ") + population = set(loaded["routed"]) + if len(population) != 256: + raise ValueError(f"expected 256 paired prompts, found {len(population)}") + truth, _, agreement = validate_assignment( + args.groups_root.resolve(), args.confusion_json.resolve(), population + ) + + by_condition: dict[str, list[np.ndarray]] = {} + for condition in ["routed", "domain_assigned"]: + arrays: list[np.ndarray] = [] + for domain_id in range(5): + keys = sorted(digest for digest, value in truth.items() if value == domain_id) + arrays.append( + np.asarray([loaded[condition][digest] for digest in keys], dtype=np.float64) + ) + by_condition[condition] = arrays + rng = np.random.default_rng(args.seed) + comparison = _summarize_pair( + by_condition["routed"], + by_condition["domain_assigned"], + DOMAINS, + rng, + args.replicates, + args.chunk_size, + ) + result = { + "analysis": "b2_selected_epoch5_router_paired_acceptance_length", + "created_utc": datetime.now(timezone.utc).isoformat(), + "numpy_version": np.__version__, + "expert_bank_checkpoint": str(args.expert_bank.resolve()), + "selected_router_checkpoint": str(args.selected_router.resolve()), + "standalone_exports_root": str(args.exports_root.resolve()), + "server_random_seed": args.server_random_seed, + "identity_artifact": str(args.identity_artifact.resolve()), + "identity_artifact_sha256": _file_sha256(args.identity_artifact.resolve()), + "groups_root": str(args.groups_root.resolve()), + "bench_root": str(args.bench_root.resolve()), + "prompt_population": { + "n_prompts": len(population), + "canonical_source_row_digest_set_sha256": digest_sequence(sorted(population)), + }, + "assignment_agreement": agreement, + "bootstrap": { + "replicates": args.replicates, + "seed": args.seed, + "chunk_size": args.chunk_size, + "interval": "two-sided 95% percentile", + "resampling": "paired by canonical source-row digest and stratified by offline assignment", + "estimator": "sum(completion_tokens) / sum(spec_verify_ct) in every replicate", + }, + "conditions": { + condition: condition_summary(by_condition[condition]) + for condition in ["routed", "domain_assigned"] + }, + "comparison": { + "left": "routed", + "right": "domain_assigned", + **comparison, + }, + "validated_sidecar_provenance": sidecar_provenance, + "limitations": [ + "The selected router checkpoint and assignment agreement were selected on this same 256-prompt assignment set.", + "Bootstrap intervals quantify evaluation-prompt uncertainty, not training-seed uncertainty.", + "The expert bank is a best-observed checkpoint selected by code-domain AL.", + ], + } + output.parent.mkdir(parents=True, exist_ok=True) + temp = output.with_suffix(output.suffix + ".tmp") + temp.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + os.replace(temp, output) + print(f"wrote {output}") + + +if __name__ == "__main__": + main() diff --git a/verification/v1/bootstrap_paired_al.py b/verification/v1/bootstrap_paired_al.py new file mode 100644 index 0000000000000000000000000000000000000000..9b314f92298276584c955339ee14ada2cb66817a --- /dev/null +++ b/verification/v1/bootstrap_paired_al.py @@ -0,0 +1,738 @@ +#!/usr/bin/env python3 +"""Paired bootstrap confidence intervals for DFlash acceptance length. + +The aggregate acceptance length (AL) is always recomputed as +``sum(completion_tokens) / sum(spec_verify_ct)`` inside each bootstrap +replicate. The script never averages per-prompt AL values. + +Two input layouts are supported: + +``aligned`` + Two checkpoint evaluations with one sidecar per domain. Rows are paired + by ``(run_idx, prompt_digest)`` and must retain identical ``prompt_idx`` + order. This is the analysis intended for the exact 3.2M + generalist-versus-MoS comparison. + +``router`` + Routed, domain-assigned (``oracle``), and ``gen`` reference evaluations + whose prompts were bucketed differently. Rows are paired by a SHA-256 digest of + the canonical source JSONL row, after joining each sidecar back to its input group in + input order. Per-domain results are then regrouped by the offline + assignment, not by the router prediction. + +Examples: + + python experiments/dflash/scripts/bootstrap_paired_al.py aligned \ + --manifest paper/submission/evidence/b1_exact_3p2m_manifest.json \ + --output /tmp/b1_bootstrap.json + + python experiments/dflash/scripts/bootstrap_paired_al.py router \ + --groups-root experiments/dflash/router_intr \ + --bench-root model/eval/routed_al_intr \ + --confusion-json experiments/dflash/router_intr/confusion.json \ + --output /tmp/router_bootstrap.json +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +from pathlib import Path +from typing import Any, Iterable + +import numpy as np + + +DEFAULT_DOMAINS = ["code", "math", "factual_qa", "creative_writing", "general"] + + +def _read_jsonl(path: Path) -> list[dict[str, Any]]: + rows: list[dict[str, Any]] = [] + with path.open() as handle: + for line_no, line in enumerate(handle, 1): + line = line.strip() + if not line: + continue + row = json.loads(line) + if not isinstance(row, dict): + raise ValueError(f"{path}:{line_no}: expected a JSON object") + rows.append(row) + return rows + + +def _file_sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for block in iter(lambda: handle.read(1024 * 1024), b""): + digest.update(block) + return digest.hexdigest() + + +def _canonical_digest(row: dict[str, Any]) -> str: + payload = json.dumps(row, sort_keys=True, ensure_ascii=False, separators=(",", ":")) + return hashlib.sha256(payload.encode("utf-8")).hexdigest() + + +def _extract_first_user(conversations: list[dict[str, Any]]) -> str | None: + """Mirror ``benchmarker.customjsonl._extract_first_user`` exactly.""" + for turn in conversations: + if turn.get("role") == "user": + return turn.get("content", "") + for turn in conversations: + if turn.get("role") != "system": + return turn.get("content", "") + return None + + +def _loaded_customjsonl_digests(path: Path, num_samples: int) -> list[str]: + """Recreate the canonical question sequence actually loaded by CustomJsonlBenchmarker.""" + digests: list[str] = [] + for line_no, row in enumerate(_read_jsonl(path), 1): + conversations = row.get("conversations", []) + if not isinstance(conversations, list): + raise ValueError(f"{path}:{line_no}: conversations must be a list") + question = _extract_first_user(conversations) + if not question: + continue + digests.append(_canonical_digest({"question": question})) + if len(digests) >= num_samples: + break + return digests + + +def _digest_sequence_sha256(digests: list[str]) -> str: + return hashlib.sha256(("\n".join(digests) + "\n").encode("ascii")).hexdigest() + + +def _resolve_one(pattern: str, base: Path) -> Path: + candidate = Path(pattern) + if not candidate.is_absolute(): + candidate = base / candidate + if any(char in str(candidate) for char in "*?["): + matches = sorted(candidate.parent.glob(candidate.name)) + if len(matches) != 1: + raise ValueError(f"expected one sidecar for {candidate}, found {len(matches)}") + return matches[0] + if not candidate.is_file(): + raise FileNotFoundError(candidate) + return candidate + + +def _as_counts(row: dict[str, Any], source: str) -> tuple[float, float]: + try: + completion = float(row["completion_tokens"]) + verify = float(row["spec_verify_ct"]) + except (KeyError, TypeError, ValueError) as error: + raise ValueError(f"{source}: missing or invalid completion/verify count") from error + if not math.isfinite(completion) or not math.isfinite(verify): + raise ValueError(f"{source}: non-finite counts completion={completion}, verify={verify}") + if completion < 0 or verify <= 0: + raise ValueError(f"{source}: invalid counts completion={completion}, verify={verify}") + return completion, verify + + +def _ratio(counts: np.ndarray) -> float: + return float(counts[:, 0].sum() / counts[:, 1].sum()) + + +def _percentile_ci(values: np.ndarray) -> list[float]: + return [float(value) for value in np.quantile(values, [0.025, 0.975])] + + +def _bootstrap_one_group( + left: np.ndarray, + right: np.ndarray, + rng: np.random.Generator, + replicates: int, + chunk_size: int, +) -> np.ndarray: + if left.shape != right.shape or left.ndim != 2 or left.shape[1] != 2: + raise ValueError(f"count arrays must both have shape [N, 2], got {left.shape}, {right.shape}") + n_rows = left.shape[0] + if n_rows == 0: + raise ValueError("cannot bootstrap an empty group") + deltas = np.empty(replicates, dtype=np.float64) + for start in range(0, replicates, chunk_size): + width = min(chunk_size, replicates - start) + picked = rng.integers(0, n_rows, size=(width, n_rows)) + left_sample = left[picked] + right_sample = right[picked] + left_ratio = left_sample[:, :, 0].sum(axis=1) / left_sample[:, :, 1].sum(axis=1) + right_ratio = right_sample[:, :, 0].sum(axis=1) / right_sample[:, :, 1].sum(axis=1) + deltas[start : start + width] = left_ratio - right_ratio + return deltas + + +def _bootstrap_stratified( + left_by_domain: list[np.ndarray], + right_by_domain: list[np.ndarray], + rng: np.random.Generator, + replicates: int, + chunk_size: int, +) -> tuple[np.ndarray, np.ndarray]: + """Return pooled-ratio and unweighted-domain-mean delta replicates.""" + pooled = np.empty(replicates, dtype=np.float64) + domain_mean = np.empty(replicates, dtype=np.float64) + for start in range(0, replicates, chunk_size): + width = min(chunk_size, replicates - start) + left_completion = np.zeros(width, dtype=np.float64) + left_verify = np.zeros(width, dtype=np.float64) + right_completion = np.zeros(width, dtype=np.float64) + right_verify = np.zeros(width, dtype=np.float64) + per_domain_delta: list[np.ndarray] = [] + for left, right in zip(left_by_domain, right_by_domain, strict=True): + if left.shape != right.shape: + raise ValueError(f"paired domain shape mismatch: {left.shape} != {right.shape}") + n_rows = left.shape[0] + picked = rng.integers(0, n_rows, size=(width, n_rows)) + left_sample = left[picked] + right_sample = right[picked] + left_ct = left_sample[:, :, 0].sum(axis=1) + left_vc = left_sample[:, :, 1].sum(axis=1) + right_ct = right_sample[:, :, 0].sum(axis=1) + right_vc = right_sample[:, :, 1].sum(axis=1) + left_completion += left_ct + left_verify += left_vc + right_completion += right_ct + right_verify += right_vc + per_domain_delta.append(left_ct / left_vc - right_ct / right_vc) + pooled[start : start + width] = ( + left_completion / left_verify - right_completion / right_verify + ) + domain_mean[start : start + width] = np.mean(per_domain_delta, axis=0) + return pooled, domain_mean + + +def _summarize_pair( + left_by_domain: list[np.ndarray], + right_by_domain: list[np.ndarray], + domains: list[str], + rng: np.random.Generator, + replicates: int, + chunk_size: int, +) -> dict[str, Any]: + per_domain: dict[str, Any] = {} + for domain, left, right in zip(domains, left_by_domain, right_by_domain, strict=True): + samples = _bootstrap_one_group(left, right, rng, replicates, chunk_size) + per_domain[domain] = { + "n_prompts": int(left.shape[0]), + "left_al": _ratio(left), + "right_al": _ratio(right), + "delta_left_minus_right": _ratio(left) - _ratio(right), + "delta_95_percentile_ci": _percentile_ci(samples), + } + + pooled_samples, mean_samples = _bootstrap_stratified( + left_by_domain, right_by_domain, rng, replicates, chunk_size + ) + left_pooled = np.concatenate(left_by_domain, axis=0) + right_pooled = np.concatenate(right_by_domain, axis=0) + left_domain_als = [_ratio(counts) for counts in left_by_domain] + right_domain_als = [_ratio(counts) for counts in right_by_domain] + return { + "per_domain": per_domain, + "token_weighted_pooled": { + "left_al": _ratio(left_pooled), + "right_al": _ratio(right_pooled), + "delta_left_minus_right": _ratio(left_pooled) - _ratio(right_pooled), + "delta_95_percentile_ci": _percentile_ci(pooled_samples), + }, + "unweighted_domain_mean": { + "left_al": float(np.mean(left_domain_als)), + "right_al": float(np.mean(right_domain_als)), + "delta_left_minus_right": float(np.mean(left_domain_als) - np.mean(right_domain_als)), + "delta_95_percentile_ci": _percentile_ci(mean_samples), + }, + } + + +def _aligned_expected_provenance( + condition: dict[str, Any], fixed_harness: dict[str, Any], domain: str +) -> dict[str, Any]: + try: + batch_size, steps, topk, num_draft_tokens = ( + int(value) for value in str(fixed_harness["config"]).split(",") + ) + prompt_source = fixed_harness["prompt_sources"][domain] + except (KeyError, TypeError, ValueError) as error: + raise ValueError( + "fixed_harness needs config=',,,' " + "and one prompt_sources entry per domain" + ) from error + expected = { + "target_model_path": fixed_harness["target_snapshot"], + "draft_model_path": condition["served_drafts"][domain], + "source_checkpoint": condition["checkpoint"], + "algorithm": fixed_harness["algorithm"], + "batch_size": batch_size, + "steps": steps, + "topk": topk, + "num_draft_tokens": num_draft_tokens, + "attention_backend": fixed_harness["attention_backend"], + "dtype": fixed_harness["dtype"], + "trust_remote_code": fixed_harness["trust_remote_code"], + "temperature": fixed_harness["temperature"], + "max_new_tokens": fixed_harness["max_new_tokens"], + "prompt_source": prompt_source, + "bench_chat_template": fixed_harness.get("bench_chat_template"), + "bench_preformat_tokenizer": fixed_harness["bench_preformat_tokenizer"], + "thinking": fixed_harness["thinking"], + } + for seed_field in ( + "server_random_seed_requested", + "server_random_seed_effective", + ): + if seed_field in fixed_harness: + expected[seed_field] = int(fixed_harness[seed_field]) + return expected + + +def _read_aggregate_als(path: Path, fixed_harness: dict[str, Any]) -> dict[int, float]: + payload = json.loads(path.read_text()) + entries = payload.get("customjsonl") + if not isinstance(entries, list): + raise ValueError(f"{path}: missing customjsonl aggregate entries") + batch_size, steps, topk, num_draft_tokens = ( + int(value) for value in str(fixed_harness["config"]).split(",") + ) + matches = [ + entry + for entry in entries + if entry.get("batch_size") == batch_size + and entry.get("steps") == steps + and entry.get("topk") == topk + and entry.get("num_draft_tokens") == num_draft_tokens + ] + if len(matches) != 1: + raise ValueError(f"{path}: expected one aggregate entry for the locked config") + metrics = matches[0].get("metrics") + if not isinstance(metrics, list) or not metrics: + raise ValueError(f"{path}: aggregate entry has no metrics") + result: dict[int, float] = {} + for run_idx, metric in enumerate(metrics): + try: + value = float(metric["accept_length"]) + except (KeyError, TypeError, ValueError) as error: + raise ValueError(f"{path}: invalid aggregate accept_length") from error + if not math.isfinite(value): + raise ValueError(f"{path}: non-finite aggregate accept_length") + result[run_idx] = value + return result + + +def _load_aligned_condition( + condition: dict[str, Any], + domains: list[str], + manifest_dir: Path, + fixed_harness: dict[str, Any], +) -> tuple[ + dict[str, dict[tuple[int, str], tuple[float, float, int]]], + dict[str, dict[str, Any]], +]: + sidecars = condition.get("sidecars") + if not isinstance(sidecars, dict): + raise ValueError("each aligned condition needs a sidecars object") + aggregates = condition.get("aggregates") + if not isinstance(aggregates, dict): + raise ValueError("each aligned condition needs an aggregates object") + required_fields = set(fixed_harness.get("required_sidecar_fields", [])) + loaded: dict[str, dict[tuple[int, str], tuple[float, float, int]]] = {} + provenance: dict[str, dict[str, Any]] = {} + for domain in domains: + path = _resolve_one(str(sidecars[domain]), manifest_dir) + expected = _aligned_expected_provenance(condition, fixed_harness, domain) + sidecar_sha256 = _file_sha256(path) + prompt_source_sha256 = _file_sha256(Path(expected["prompt_source"])) + expected_source_sha256 = fixed_harness["prompt_source_sha256"][domain] + if prompt_source_sha256 != expected_source_sha256: + raise ValueError( + f"{expected['prompt_source']}: sha256={prompt_source_sha256}, " + f"expected locked digest {expected_source_sha256}" + ) + expected_count = int(fixed_harness["expected_prompt_counts"][domain]) + expected_prompt_digests = _loaded_customjsonl_digests( + Path(expected["prompt_source"]), expected_count + ) + if len(expected_prompt_digests) != expected_count: + raise ValueError( + f"{expected['prompt_source']}: benchmark loader yields " + f"{len(expected_prompt_digests)} prompts, expected {expected_count}" + ) + keyed: dict[tuple[int, str], tuple[float, float, int]] = {} + observed_provenance: dict[str, Any] | None = None + for line_no, row in enumerate(_read_jsonl(path), 1): + missing = sorted(required_fields - set(row)) + if missing: + raise ValueError( + f"{path}:{line_no}: required sidecar fields missing: {missing}" + ) + for field, expected_value in expected.items(): + if row.get(field) != expected_value: + raise ValueError( + f"{path}:{line_no}: {field}={row.get(field)!r}, " + f"expected {expected_value!r}" + ) + current_provenance = { + **expected, + "sidecar": str(path), + "sidecar_sha256": sidecar_sha256, + "prompt_source_sha256": prompt_source_sha256, + } + if observed_provenance is None: + observed_provenance = current_provenance + elif current_provenance != observed_provenance: + raise ValueError(f"{path}:{line_no}: sidecar provenance changes within one cell") + digest = str(row["prompt_digest"]) + if len(digest) != 64 or any(char not in "0123456789abcdef" for char in digest): + raise ValueError(f"{path}:{line_no}: invalid prompt_digest {digest!r}") + run_idx = int(row.get("run_idx", 0)) + prompt_idx = int(row["prompt_idx"]) + key = (run_idx, digest) + if key in keyed: + raise ValueError(f"{path}:{line_no}: duplicate pairing key {key}") + keyed[key] = (*_as_counts(row, f"{path}:{line_no}"), prompt_idx) + if observed_provenance is None: + raise ValueError(f"{path}: empty sidecar") + aggregate_path = _resolve_one(str(aggregates[domain]), manifest_dir) + aggregate_als = _read_aggregate_als(aggregate_path, fixed_harness) + run_indices = sorted({key[0] for key in keyed}) + expected_run_indices = list(fixed_harness.get("expected_run_indices", [0])) + if run_indices != expected_run_indices: + raise ValueError( + f"{path}: run indices {run_indices} != expected {expected_run_indices}" + ) + for run_idx in run_indices: + indices = sorted(value[2] for key, value in keyed.items() if key[0] == run_idx) + if len(indices) != expected_count: + raise ValueError( + f"{path}: run {run_idx} has {len(indices)} prompts, expected {expected_count}" + ) + if indices != list(range(len(indices))): + raise ValueError( + f"{path}: run {run_idx} prompt_idx values are not contiguous from zero" + ) + observed_prompt_digests = [ + key[1] + for key, value in sorted( + ( + (key, value) + for key, value in keyed.items() + if key[0] == run_idx + ), + key=lambda item: item[1][2], + ) + ] + if observed_prompt_digests != expected_prompt_digests: + mismatch = next( + ( + index + for index, (observed, expected_digest) in enumerate( + zip( + observed_prompt_digests, + expected_prompt_digests, + strict=True, + ) + ) + if observed != expected_digest + ), + None, + ) + raise ValueError( + f"{path}: run {run_idx} prompt digests do not match the questions " + f"loaded from {expected['prompt_source']} (first mismatch={mismatch})" + ) + counts = np.asarray( + [value[:2] for key, value in keyed.items() if key[0] == run_idx], + dtype=np.float64, + ) + if run_idx not in aggregate_als or not math.isclose( + _ratio(counts), aggregate_als[run_idx], rel_tol=0.0, abs_tol=1e-10 + ): + raise ValueError( + f"{path}: pooled sidecar AL {_ratio(counts)} disagrees with " + f"aggregate run {run_idx} value {aggregate_als.get(run_idx)}" + ) + loaded[domain] = keyed + observed_provenance["aggregate_result"] = str(aggregate_path) + observed_provenance["aggregate_result_sha256"] = _file_sha256(aggregate_path) + observed_provenance["aggregate_accept_length"] = aggregate_als + observed_provenance["loaded_prompt_digest_sequence_sha256"] = ( + _digest_sequence_sha256(expected_prompt_digests) + ) + provenance[domain] = observed_provenance + return loaded, provenance + + +def _run_aligned(args: argparse.Namespace) -> dict[str, Any]: + manifest_path = Path(args.manifest).resolve() + manifest = json.loads(manifest_path.read_text()) + domains = list(manifest.get("domains", DEFAULT_DOMAINS)) + left_spec = manifest["left"] + right_spec = manifest["right"] + fixed_harness = manifest["fixed_harness"] + left, left_provenance = _load_aligned_condition( + left_spec, domains, manifest_path.parent, fixed_harness + ) + right, right_provenance = _load_aligned_condition( + right_spec, domains, manifest_path.parent, fixed_harness + ) + left_arrays: list[np.ndarray] = [] + right_arrays: list[np.ndarray] = [] + for domain in domains: + if set(left[domain]) != set(right[domain]): + missing_left = sorted(set(right[domain]) - set(left[domain]))[:5] + missing_right = sorted(set(left[domain]) - set(right[domain]))[:5] + raise ValueError( + f"{domain}: prompt keys differ; missing left={missing_left}, missing right={missing_right}" + ) + keys = sorted(left[domain]) + for key in keys: + if left[domain][key][2] != right[domain][key][2]: + raise ValueError( + f"{domain}: prompt order differs for run/digest {key}: " + f"{left[domain][key][2]} != {right[domain][key][2]}" + ) + left_arrays.append( + np.asarray([left[domain][key][:2] for key in keys], dtype=np.float64) + ) + right_arrays.append( + np.asarray([right[domain][key][:2] for key in keys], dtype=np.float64) + ) + rng = np.random.default_rng(args.seed) + return { + "analysis": "aligned_paired_acceptance_length", + "manifest": str(manifest_path), + "manifest_sha256": _file_sha256(manifest_path), + "numpy_version": np.__version__, + "left": str(left_spec.get("label", "left")), + "right": str(right_spec.get("label", "right")), + "domains": domains, + "validated_sidecar_provenance": { + "left": left_provenance, + "right": right_provenance, + }, + "bootstrap": { + "replicates": args.replicates, + "seed": args.seed, + "chunk_size": args.chunk_size, + "interval": "two-sided 95% percentile", + "resampling": "paired within domain; domains retained at observed sizes", + }, + "results": _summarize_pair( + left_arrays, right_arrays, domains, rng, args.replicates, args.chunk_size + ), + } + + +def _find_sidecar(cell: Path) -> Path: + matches = sorted(cell.glob("*perprompt*.jsonl")) + if len(matches) != 1: + raise ValueError(f"expected one per-prompt sidecar under {cell}, found {len(matches)}") + return matches[0] + + +def _load_grouped_condition( + condition: str, groups_root: Path, bench_root: Path +) -> dict[str, tuple[float, float]]: + group_dir = groups_root / f"{condition}_groups" + joined: dict[str, tuple[float, float]] = {} + for domain_id in range(5): + source_rows = _read_jsonl(group_dir / f"{domain_id}.jsonl") + metric_path = _find_sidecar(bench_root / condition / str(domain_id)) + metric_rows = _read_jsonl(metric_path) + if len(source_rows) != len(metric_rows): + raise ValueError( + f"{condition}/{domain_id}: {len(source_rows)} input rows != " + f"{len(metric_rows)} sidecar rows" + ) + for row_no, (source, metric) in enumerate(zip(source_rows, metric_rows, strict=True), 1): + digest = _canonical_digest(source) + if digest in joined: + raise ValueError(f"{condition}/{domain_id}:{row_no}: duplicate source-row digest") + joined[digest] = _as_counts(metric, f"{metric_path}:{row_no}") + return joined + + +def _run_router(args: argparse.Namespace) -> dict[str, Any]: + groups_root = Path(args.groups_root).resolve() + bench_root = Path(args.bench_root).resolve() + conditions = ["routed", "oracle", "gen"] + loaded = { + condition: _load_grouped_condition(condition, groups_root, bench_root) + for condition in conditions + } + prompt_sets = [set(rows) for rows in loaded.values()] + if not all(prompt_set == prompt_sets[0] for prompt_set in prompt_sets[1:]): + raise ValueError("routed, oracle, and gen-condition prompt populations differ") + + true_domain: dict[str, int] = {} + for domain_id in range(5): + for row in _read_jsonl(groups_root / "oracle_groups" / f"{domain_id}.jsonl"): + digest = _canonical_digest(row) + if digest in true_domain: + raise ValueError(f"duplicate oracle source-row digest {digest}") + true_domain[digest] = domain_id + if set(true_domain) != prompt_sets[0]: + raise ValueError("offline-assignment rows do not match evaluated prompt population") + + predicted_domain: dict[str, int] = {} + for domain_id in range(5): + for row in _read_jsonl(groups_root / "routed_groups" / f"{domain_id}.jsonl"): + digest = _canonical_digest(row) + if digest in predicted_domain: + raise ValueError(f"duplicate routed source-row digest {digest}") + predicted_domain[digest] = domain_id + if set(predicted_domain) != prompt_sets[0]: + raise ValueError("router-prediction rows do not match evaluated prompt population") + computed_confusion = [[0 for _ in range(5)] for _ in range(5)] + for digest, true_id in true_domain.items(): + computed_confusion[true_id][predicted_domain[digest]] += 1 + computed_agreement = sum( + computed_confusion[index][index] for index in range(5) + ) / len(true_domain) + computed_recall = [ + computed_confusion[index][index] / sum(computed_confusion[index]) + for index in range(5) + ] + + domains = list(DEFAULT_DOMAINS) + by_condition: dict[str, list[np.ndarray]] = {} + for condition in conditions: + domain_arrays: list[np.ndarray] = [] + for domain_id in range(5): + keys = sorted(key for key, value in true_domain.items() if value == domain_id) + domain_arrays.append( + np.asarray([loaded[condition][key] for key in keys], dtype=np.float64) + ) + by_condition[condition] = domain_arrays + + comparison_specs = [ + ("routed_minus_domain_assigned", "routed", "oracle"), + ("routed_minus_gen_condition", "routed", "gen"), + ("domain_assigned_minus_gen_condition", "oracle", "gen"), + ] + rng = np.random.default_rng(args.seed) + comparisons = { + name: { + "left": left, + "right": right, + **_summarize_pair( + by_condition[left], + by_condition[right], + domains, + rng, + args.replicates, + args.chunk_size, + ), + } + for name, left, right in comparison_specs + } + + condition_results: dict[str, Any] = {} + for condition, arrays in by_condition.items(): + domain_als = [_ratio(counts) for counts in arrays] + condition_results[condition] = { + "n_prompts": int(sum(counts.shape[0] for counts in arrays)), + "token_weighted_pooled_al": _ratio(np.concatenate(arrays, axis=0)), + "unweighted_domain_mean_al": float(np.mean(domain_als)), + "per_domain_al": dict(zip(domains, domain_als, strict=True)), + } + + agreement = None + if args.confusion_json: + confusion_path = Path(args.confusion_json).resolve() + confusion = json.loads(confusion_path.read_text()) + if confusion.get("confusion") != computed_confusion: + raise ValueError( + f"{confusion_path}: confusion matrix disagrees with routed/oracle group membership" + ) + if not math.isclose( + float(confusion.get("overall_accuracy")), computed_agreement, abs_tol=1e-12 + ): + raise ValueError(f"{confusion_path}: overall agreement disagrees with confusion matrix") + artifact_recall = [float(value) for value in confusion.get("per_domain_recall", [])] + if len(artifact_recall) != 5 or not np.allclose( + artifact_recall, computed_recall, atol=1e-12, rtol=0.0 + ): + raise ValueError(f"{confusion_path}: per-domain recall disagrees with confusion matrix") + agreement = { + "artifact": args.confusion_artifact or str(confusion_path), + "validated_against_group_membership": True, + "overall_assignment_agreement": computed_agreement, + "confusion": computed_confusion, + "per_domain_recall": computed_recall, + } + + return { + "analysis": "router_paired_acceptance_length", + "numpy_version": np.__version__, + "router_training_checkpoint": args.router_training_checkpoint, + "selected_router_checkpoint": args.selected_router_checkpoint, + "drafter_exports_checkpoint": args.drafter_exports_checkpoint, + "drafter_export_provenance": args.drafter_export_provenance, + "groups_root": args.groups_artifact or str(groups_root), + "bench_root": args.bench_artifact or str(bench_root), + "domains": domains, + "bootstrap": { + "replicates": args.replicates, + "seed": args.seed, + "chunk_size": args.chunk_size, + "interval": "two-sided 95% percentile", + "resampling": "paired by canonical source-row digest and stratified by offline assignment", + }, + "assignment_agreement": agreement, + "conditions": condition_results, + "comparisons": comparisons, + } + + +def _write_output(result: dict[str, Any], output: str | None) -> None: + rendered = json.dumps(result, indent=2, sort_keys=True) + "\n" + if output: + path = Path(output) + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(rendered) + print(f"wrote {path.resolve()}") + else: + print(rendered, end="") + + +def _add_bootstrap_options(parser: argparse.ArgumentParser) -> None: + parser.add_argument("--replicates", type=int, default=50_000) + parser.add_argument("--seed", type=int, default=20_260_719) + parser.add_argument("--chunk-size", type=int, default=500) + parser.add_argument("--output") + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + subparsers = parser.add_subparsers(dest="command", required=True) + + aligned = subparsers.add_parser("aligned", help="paired sidecars with stable prompt_idx") + aligned.add_argument("--manifest", required=True) + _add_bootstrap_options(aligned) + + router = subparsers.add_parser("router", help="join differently bucketed router evaluations") + router.add_argument("--groups-root", required=True) + router.add_argument("--bench-root", required=True) + router.add_argument("--confusion-json") + router.add_argument("--groups-artifact", help="stable provenance path recorded in output") + router.add_argument("--bench-artifact", help="stable provenance path recorded in output") + router.add_argument("--confusion-artifact", help="stable provenance path recorded in output") + router.add_argument("--router-training-checkpoint") + router.add_argument("--selected-router-checkpoint") + router.add_argument("--drafter-exports-checkpoint") + router.add_argument("--drafter-export-provenance") + _add_bootstrap_options(router) + + args = parser.parse_args() + if args.replicates < 1 or args.chunk_size < 1: + parser.error("replicates and chunk-size must be positive") + result = _run_aligned(args) if args.command == "aligned" else _run_router(args) + _write_output(result, args.output) + + +if __name__ == "__main__": + main() diff --git a/verification/v1/r1_mainfig_verify_exports.py b/verification/v1/r1_mainfig_verify_exports.py new file mode 100644 index 0000000000000000000000000000000000000000..bc5bca028550b5a88f83e56b6ebb6f6d665b4b3c --- /dev/null +++ b/verification/v1/r1_mainfig_verify_exports.py @@ -0,0 +1,175 @@ +#!/usr/bin/env python3 +"""Tensor-for-tensor identity audit for R1 standalone MLP-selection exports. + +Generalizes the b2 verifier's compare_exports to any MoS bank / export root / +domain subset: for each requested domain it proves the exported num_domains=1 +draft is exactly the bank's shared parameters plus the bank's +``domain_mlps.`` tensors remapped to ``.mlp.``, with no extra tensors. + +Usage: + r1_mainfig_verify_exports.py --bank \ + --exports-root \ + --domains code [math ...] --output +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import os +import sys +from contextlib import ExitStack +from datetime import datetime, timezone +from pathlib import Path + +import torch +from safetensors import safe_open + +DOMAIN_ID = {"code": 0, "math": 1, "factual_qa": 2, "creative_writing": 3, "general": 4} +PARITY_KEYS = [ + "hidden_size", + "intermediate_size", + "num_attention_heads", + "num_key_value_heads", + "head_dim", + "rms_norm_eps", + "rope_theta", + "vocab_size", + "num_hidden_layers", + "dflash_config", + "block_size", +] +REMOTE_FILES = ["dflash.py", "modeling_dflash.py", "utils.py"] + + +def file_sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for block in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(block) + return digest.hexdigest() + + +class ShardedCheckpoint: + def __init__(self, root: Path, stack: ExitStack): + self.root = root.resolve() + index_path = self.root / "model.safetensors.index.json" + if index_path.is_file(): + index = json.loads(index_path.read_text()) + self.weight_map = dict(index["weight_map"]) + filenames = sorted(set(self.weight_map.values())) + else: + files = sorted(self.root.glob("*.safetensors")) + if len(files) != 1: + raise ValueError(f"{self.root}: expected an index or one safetensors file") + filenames = [files[0].name] + with safe_open(files[0], framework="pt", device="cpu") as handle: + self.weight_map = {key: files[0].name for key in handle.keys()} + self.handles = { + name: stack.enter_context( + safe_open(self.root / name, framework="pt", device="cpu") + ) + for name in filenames + } + + def tensor(self, key: str) -> torch.Tensor: + if key not in self.weight_map: + raise KeyError(f"{self.root}: missing tensor {key}") + return self.handles[self.weight_map[key]].get_tensor(key) + + +def verify_domain(bank: ShardedCheckpoint, bank_config: dict, export_dir: Path, domain: str) -> dict: + domain_id = DOMAIN_ID[domain] + export_path = export_dir / "model.safetensors" + config_path = export_dir / "config.json" + if not export_path.is_file() or not config_path.is_file(): + raise FileNotFoundError(f"incomplete export: {export_dir}") + export_config = json.loads(config_path.read_text()) + if int(export_config.get("num_domains", 1)) != 1: + raise ValueError(f"{config_path}: num_domains must be 1") + for key in PARITY_KEYS: + if export_config.get(key) != bank_config.get(key): + raise ValueError( + f"{config_path}: {key}={export_config.get(key)!r} != bank {bank_config.get(key)!r}" + ) + missing_remote = [name for name in REMOTE_FILES if not (export_dir / name).is_file()] + if missing_remote: + raise ValueError(f"{export_dir}: missing remote-code files {missing_remote}") + compared = 0 + mapped_mlp = 0 + with safe_open(export_path, framework="pt", device="cpu") as exported: + export_keys = list(exported.keys()) + if len(export_keys) != 58: + raise ValueError(f"{export_path}: expected 58 tensors, found {len(export_keys)}") + if any(k.startswith("router_head.") for k in export_keys): + raise ValueError(f"{export_path}: router_head tensors must not be exported") + for export_key in export_keys: + if ".mlp." in export_key: + source_key = export_key.replace(".mlp.", f".domain_mlps.{domain_id}.", 1) + mapped_mlp += 1 + else: + source_key = export_key + if not torch.equal(exported.get_tensor(export_key), bank.tensor(source_key)): + raise ValueError( + f"{export_path}: {export_key} differs from {bank.root}:{source_key}" + ) + compared += 1 + if mapped_mlp != 15: + raise ValueError(f"{export_path}: expected 15 mapped MLP tensors, found {mapped_mlp}") + return { + "domain_id": domain_id, + "export_dir": str(export_dir.resolve()), + "model_sha256": file_sha256(export_path), + "config_sha256": file_sha256(config_path), + "tensor_equality": { + "validated": True, + "compared_tensors": compared, + "mapped_domain_mlp_tensors": mapped_mlp, + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--bank", required=True, type=Path) + parser.add_argument("--exports-root", required=True, type=Path) + parser.add_argument("--domains", nargs="+", required=True, choices=sorted(DOMAIN_ID)) + parser.add_argument("--output", required=True, type=Path) + args = parser.parse_args() + + output = args.output.resolve() + if output.exists(): + raise FileExistsError(f"refusing to overwrite {output}") + + bank_config = json.loads((args.bank / "config.json").read_text()) + if int(bank_config.get("num_domains", 0)) != 5: + raise ValueError(f"{args.bank}: num_domains={bank_config.get('num_domains')}, expected 5") + + with ExitStack() as stack: + bank = ShardedCheckpoint(args.bank, stack) + validation = {} + for domain in args.domains: + validation[domain] = verify_domain( + bank, bank_config, args.exports_root / domain, domain + ) + print(f"[verify {domain}] OK: {validation[domain]['tensor_equality']}") + + result = { + "analysis": "r1_mainfig_export_identity", + "created_utc": datetime.now(timezone.utc).isoformat(), + "torch_version": torch.__version__, + "bank": str(args.bank.resolve()), + "exports_root": str(args.exports_root.resolve()), + "domain_mapping": {d: DOMAIN_ID[d] for d in args.domains}, + "validation": validation, + } + output.parent.mkdir(parents=True, exist_ok=True) + temp = output.with_suffix(".tmp") + temp.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + os.replace(temp, output) + print(f"wrote {output}") + + +if __name__ == "__main__": + main() diff --git a/verification/v1/verify_b1_exact_3p2m_routed_artifacts.py b/verification/v1/verify_b1_exact_3p2m_routed_artifacts.py new file mode 100644 index 0000000000000000000000000000000000000000..276f60d3fcdc78173f6e8b127326db28931092e4 --- /dev/null +++ b/verification/v1/verify_b1_exact_3p2m_routed_artifacts.py @@ -0,0 +1,244 @@ +#!/usr/bin/env python3 +"""Fail-closed artifact audit for routed evaluation of the exact 3.2M MoS bank. + +The selected mean-only sidecar consumes frozen target-model prompt features, so +its request assignments can be applied to another five-domain MoS bank with the +same domain-index mapping. This audit verifies the selected head and groups, +proves tensor equality between the exact matched-volume bank and its five served +exports, and proves that the assignment population is exactly the B1 population. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import os +from contextlib import ExitStack +from datetime import datetime, timezone +from pathlib import Path +from typing import Any + +import torch + +from verify_b2_epoch5_artifacts import ( + DOMAINS, + ShardedCheckpoint, + canonical_row, + compare_exports, + digest_sequence, + file_sha256, + read_jsonl, + row_digest, + tensor_sha256, + validate_groups, +) + + +EXPECTED_ROUTER_KEYS = { + "router_head.0.weight", + "router_head.0.bias", + "router_head.1.weight", + "router_head.1.bias", + "router_head.4.weight", + "router_head.4.bias", +} + + +def loaded_question(row: dict[str, Any], source: Path) -> str: + conversations = row.get("conversations", []) + if not isinstance(conversations, list): + raise ValueError(f"{source}: conversations must be a list") + for turn in conversations: + if turn.get("role") == "user": + question = turn.get("content", "") + if question: + return question + for turn in conversations: + if turn.get("role") != "system": + question = turn.get("content", "") + if question: + return question + raise ValueError(f"{source}: row would be skipped by CustomJsonlBenchmarker") + + +def question_digest(row: dict[str, Any], source: Path) -> str: + payload = json.dumps( + {"question": loaded_question(row, source)}, + sort_keys=True, + ensure_ascii=False, + separators=(",", ":"), + ) + return hashlib.sha256(payload.encode("utf-8")).hexdigest() + + +def validate_router_head(selected_router: ShardedCheckpoint) -> dict[str, Any]: + keys = sorted( + key for key in selected_router.weight_map if key.startswith("router_head.") + ) + if set(keys) != EXPECTED_ROUTER_KEYS: + raise ValueError(f"unexpected router head keys: {keys}") + weight = selected_router.tensor("router_head.1.weight") + output = selected_router.tensor("router_head.4.weight") + if tuple(weight.shape) != (512, 20_480): + raise ValueError(f"unexpected router projection shape: {tuple(weight.shape)}") + if tuple(output.shape) != (5, 512): + raise ValueError(f"unexpected router classifier shape: {tuple(output.shape)}") + return { + "architecture": "LayerNorm(20480)-Linear(20480,512)-GELU-Dropout(0.2)-Linear(512,5)", + "head_tensor_sha256": { + key: tensor_sha256(selected_router.tensor(key)) for key in keys + }, + } + + +def validate_b1_population( + data_root: Path, assignment_set: dict[str, Any] +) -> dict[str, Any]: + source_files: dict[str, Any] = {} + b1_population: list[str] = [] + assignment_population: list[str] = [] + for domain_id, domain in enumerate(DOMAINS): + source = (data_root / domain / "test.jsonl").resolve() + oracle = Path( + assignment_set["groups"]["domain_assigned"][ + ["code", "math", "factualqa", "creativewriting", "general"][domain_id] + ]["path"] + ).resolve() + source_rows = read_jsonl(source) + oracle_rows = read_jsonl(oracle) + source_questions = [question_digest(row, source) for row in source_rows] + oracle_questions = [question_digest(row, oracle) for row in oracle_rows] + if len(source_questions) != len(set(source_questions)): + raise ValueError(f"{source}: duplicate loaded questions") + if len(oracle_questions) != len(set(oracle_questions)): + raise ValueError(f"{oracle}: duplicate loaded questions") + if set(source_questions) != set(oracle_questions): + raise ValueError( + f"{source} and {oracle} do not contain the same loaded questions" + ) + source_rows_digest = [row_digest(row) for row in source_rows] + oracle_rows_digest = [row_digest(row) for row in oracle_rows] + b1_population.extend(source_questions) + assignment_population.extend(oracle_questions) + source_files[domain] = { + "path": str(source), + "sha256": file_sha256(source), + "n_prompts": len(source_rows), + "loaded_question_digest_sequence_sha256": digest_sequence(source_questions), + "loaded_question_digest_set_sha256": digest_sequence( + sorted(source_questions) + ), + "canonical_source_row_digest_sequence_sha256": digest_sequence( + source_rows_digest + ), + "offline_assignment_group": str(oracle), + "offline_assignment_group_sha256": file_sha256(oracle), + "offline_assignment_loaded_question_digest_sequence_sha256": digest_sequence( + oracle_questions + ), + "offline_assignment_canonical_row_digest_sequence_sha256": digest_sequence( + oracle_rows_digest + ), + "loaded_question_set_identical_to_offline_assignment_group": True, + "full_json_rows_identical_in_order": [ + canonical_row(row) for row in source_rows + ] + == [canonical_row(row) for row in oracle_rows], + } + if len(b1_population) != len(set(b1_population)): + raise ValueError("B1 source population contains duplicate loaded questions") + if len(assignment_population) != len(set(assignment_population)): + raise ValueError("assignment population contains duplicate loaded questions") + if set(b1_population) != set(assignment_population): + raise ValueError("B1 and assignment populations differ by loaded question") + observed = digest_sequence(sorted(b1_population)) + return { + "n_prompts": len(b1_population), + "canonical_loaded_question_digest_set_sha256": observed, + "identical_to_router_assignment_population_by_loaded_question": True, + "identity_key": ( + "SHA-256 of the canonical {'question': first_user_content} object, " + "the same key serialized as prompt_digest in benchmark sidecars" + ), + "row_difference_note": ( + "B1 sources retain an id field and source order; router groups omit id and " + "follow feature-cache order. Exact question-digest sets, not full JSON rows, " + "are therefore the evaluation identity." + ), + "router_assignment_canonical_row_digest_set_sha256": assignment_set[ + "prompt_population" + ]["canonical_row_digest_set_sha256"], + "sources": source_files, + } + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--expert-bank", required=True, type=Path) + parser.add_argument("--selected-router", required=True, type=Path) + parser.add_argument("--exports-root", required=True, type=Path) + parser.add_argument("--features", required=True, type=Path) + parser.add_argument("--groups-root", required=True, type=Path) + parser.add_argument("--confusion-json", required=True, type=Path) + parser.add_argument("--b1-data-root", required=True, type=Path) + parser.add_argument("--b2-router-identity", required=True, type=Path) + parser.add_argument("--output", required=True, type=Path) + args = parser.parse_args() + + output = args.output.resolve() + if output.exists(): + raise FileExistsError(f"refusing to overwrite {output}") + prior_identity = json.loads(args.b2_router_identity.read_text()) + if Path(prior_identity["selected_router"]["path"]).resolve() != args.selected_router.resolve(): + raise ValueError("B2 identity artifact records a different selected router") + + with ExitStack() as stack: + expert_bank = ShardedCheckpoint(args.expert_bank, stack) + selected_router = ShardedCheckpoint(args.selected_router, stack) + assignment = validate_groups( + args.features.resolve(), + args.groups_root.resolve(), + args.confusion_json.resolve(), + selected_router, + ) + result = { + "analysis": "b1_exact_3p2m_unified_router_artifact_identity", + "created_utc": datetime.now(timezone.utc).isoformat(), + "torch_version": torch.__version__, + "served_expert_bank": { + "path": str(expert_bank.root), + "weight_artifact_sha256": expert_bank.artifact_hashes(), + "training_samples": 3_196_160, + }, + "standalone_exports": { + "root": str(args.exports_root.resolve()), + "validation": compare_exports(expert_bank, args.exports_root.resolve()), + }, + "selected_mean_only_sidecar": { + "path": str(selected_router.root), + "reuse_basis": ( + "The head consumes frozen target-model prompt features and uses the same " + "five domain indices; assignments are independent of served draft weights." + ), + "prior_tensor_identity_artifact": str(args.b2_router_identity.resolve()), + "prior_tensor_identity_artifact_sha256": file_sha256( + args.b2_router_identity.resolve() + ), + **validate_router_head(selected_router), + }, + "assignment_set": assignment, + "b1_evaluation_population": validate_b1_population( + args.b1_data_root.resolve(), assignment + ), + } + + output.parent.mkdir(parents=True, exist_ok=True) + temp = output.with_suffix(output.suffix + ".tmp") + temp.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + os.replace(temp, output) + print(f"wrote {output}") + + +if __name__ == "__main__": + main() diff --git a/verification/v1/verify_b2_epoch5_artifacts.py b/verification/v1/verify_b2_epoch5_artifacts.py new file mode 100644 index 0000000000000000000000000000000000000000..ba1f36243d18098f2150350eb824f98dddabbfad --- /dev/null +++ b/verification/v1/verify_b2_epoch5_artifacts.py @@ -0,0 +1,419 @@ +#!/usr/bin/env python3 +"""Fail-closed identity audit for the selected epoch-5 router evaluation. + +This audit keeps three artifacts separate: + +* the selected G-init. MoS expert bank (``epoch_5_step_149820``); +* the detached router checkpoint selected on the 256-row assignment set; and +* the five standalone exports served by the generation harness. + +It proves tensor-for-tensor that the standalone exports came from the selected +expert bank, proves that all non-router tensors in the detached-router +checkpoint are unchanged from that bank, and recomputes the selected router's +predictions and assignment agreement from the retained target features. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import os +from contextlib import ExitStack +from datetime import datetime, timezone +from pathlib import Path +from typing import Any + +import torch +import torch.nn.functional as F +from safetensors import safe_open + + +DOMAINS = ["code", "math", "factual_qa", "creative_writing", "general"] +GROUP_DOMAINS = ["code", "math", "factualqa", "creativewriting", "general"] +ROUTER_FEATURE_DIM = 20_480 + + +def file_sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for block in iter(lambda: handle.read(4 * 1024 * 1024), b""): + digest.update(block) + return digest.hexdigest() + + +def canonical_row(row: dict[str, Any]) -> str: + return json.dumps(row, sort_keys=True, ensure_ascii=False, separators=(",", ":")) + + +def row_digest(row: dict[str, Any]) -> str: + return hashlib.sha256(canonical_row(row).encode("utf-8")).hexdigest() + + +def digest_sequence(values: list[str]) -> str: + return hashlib.sha256(("\n".join(values) + "\n").encode("ascii")).hexdigest() + + +def read_jsonl(path: Path) -> list[dict[str, Any]]: + rows: list[dict[str, Any]] = [] + with path.open() as handle: + for line_no, line in enumerate(handle, 1): + if not line.strip(): + continue + value = json.loads(line) + if not isinstance(value, dict): + raise ValueError(f"{path}:{line_no}: expected JSON object") + rows.append(value) + return rows + + +class ShardedCheckpoint: + def __init__(self, root: Path, stack: ExitStack): + self.root = root.resolve() + index_path = self.root / "model.safetensors.index.json" + if index_path.is_file(): + index = json.loads(index_path.read_text()) + self.weight_map = dict(index["weight_map"]) + filenames = sorted(set(self.weight_map.values())) + else: + files = sorted(self.root.glob("*.safetensors")) + if len(files) != 1: + raise ValueError(f"{self.root}: expected an index or one safetensors file") + filenames = [files[0].name] + with safe_open(files[0], framework="pt", device="cpu") as handle: + self.weight_map = {key: files[0].name for key in handle.keys()} + self.handles = { + name: stack.enter_context( + safe_open(self.root / name, framework="pt", device="cpu") + ) + for name in filenames + } + self.files = filenames + + def tensor(self, key: str) -> torch.Tensor: + if key not in self.weight_map: + raise KeyError(f"{self.root}: missing tensor {key}") + return self.handles[self.weight_map[key]].get_tensor(key) + + def artifact_hashes(self) -> dict[str, str]: + paths = [self.root / name for name in self.files] + index = self.root / "model.safetensors.index.json" + if index.is_file(): + paths.append(index) + return {path.name: file_sha256(path) for path in paths} + + +def tensor_sha256(tensor: torch.Tensor) -> str: + value = tensor.detach().cpu().contiguous() + digest = hashlib.sha256() + digest.update(str(value.dtype).encode("ascii")) + digest.update(json.dumps(list(value.shape)).encode("ascii")) + digest.update(value.view(torch.uint8).numpy().tobytes()) + return digest.hexdigest() + + +def compare_exports( + expert_bank: ShardedCheckpoint, exports_root: Path +) -> dict[str, Any]: + source_config = json.loads((expert_bank.root / "config.json").read_text()) + if int(source_config.get("num_domains", 0)) != 5: + raise ValueError(f"expert bank num_domains={source_config.get('num_domains')}, expected 5") + parity_keys = [ + "hidden_size", + "intermediate_size", + "num_attention_heads", + "num_key_value_heads", + "head_dim", + "rms_norm_eps", + "rope_theta", + "vocab_size", + "num_hidden_layers", + "dflash_config", + "block_size", + ] + result: dict[str, Any] = {} + for domain_id, domain in enumerate(DOMAINS): + export_dir = (exports_root / domain).resolve() + export_path = export_dir / "model.safetensors" + config_path = export_dir / "config.json" + if not export_path.is_file() or not config_path.is_file(): + raise FileNotFoundError(f"incomplete export: {export_dir}") + export_config = json.loads(config_path.read_text()) + if int(export_config.get("num_domains", 1)) != 1: + raise ValueError(f"{config_path}: num_domains must be 1") + for key in parity_keys: + if export_config.get(key) != source_config.get(key): + raise ValueError( + f"{config_path}: {key}={export_config.get(key)!r}, " + f"source={source_config.get(key)!r}" + ) + compared = 0 + mapped_mlp = 0 + with safe_open(export_path, framework="pt", device="cpu") as exported: + export_keys = list(exported.keys()) + if len(export_keys) != 58: + raise ValueError(f"{export_path}: expected 58 tensors, found {len(export_keys)}") + for export_key in export_keys: + if ".mlp." in export_key: + source_key = export_key.replace( + ".mlp.", f".domain_mlps.{domain_id}.", 1 + ) + mapped_mlp += 1 + else: + source_key = export_key + if not torch.equal(exported.get_tensor(export_key), expert_bank.tensor(source_key)): + raise ValueError( + f"{export_path}: {export_key} differs from {expert_bank.root}:{source_key}" + ) + compared += 1 + if mapped_mlp != 15: + raise ValueError(f"{export_path}: expected 15 mapped MLP tensors, found {mapped_mlp}") + result[domain] = { + "domain_id": domain_id, + "export_dir": str(export_dir), + "model_sha256": file_sha256(export_path), + "config_sha256": file_sha256(config_path), + "tensor_equality": { + "validated": True, + "compared_tensors": compared, + "mapped_domain_mlp_tensors": mapped_mlp, + }, + } + return result + + +def compare_router_backbone( + expert_bank: ShardedCheckpoint, selected_router: ShardedCheckpoint +) -> dict[str, Any]: + source_keys = set(expert_bank.weight_map) + router_keys = sorted(key for key in selected_router.weight_map if key.startswith("router_head.")) + non_router_keys = set(selected_router.weight_map) - set(router_keys) + if non_router_keys != source_keys: + raise ValueError( + "selected router non-head key set differs from expert bank: " + f"missing={sorted(source_keys - non_router_keys)[:5]}, " + f"extra={sorted(non_router_keys - source_keys)[:5]}" + ) + for key in sorted(source_keys): + if not torch.equal(expert_bank.tensor(key), selected_router.tensor(key)): + raise ValueError(f"selected router changed frozen expert-bank tensor {key}") + expected_router_keys = { + "router_head.0.weight", + "router_head.0.bias", + "router_head.1.weight", + "router_head.1.bias", + "router_head.4.weight", + "router_head.4.bias", + } + if set(router_keys) != expected_router_keys: + raise ValueError(f"unexpected router head keys: {router_keys}") + return { + "frozen_expert_bank_equality": { + "validated": True, + "compared_tensors": len(source_keys), + }, + "head_tensor_sha256": { + key: tensor_sha256(selected_router.tensor(key)) for key in router_keys + }, + } + + +def router_predictions( + selected_router: ShardedCheckpoint, features: torch.Tensor +) -> torch.Tensor: + if features.ndim != 2 or features.shape[1] < ROUTER_FEATURE_DIM: + raise ValueError(f"unexpected feature shape {tuple(features.shape)}") + value = features[:, :ROUTER_FEATURE_DIM].float() + value = F.layer_norm( + value, + (ROUTER_FEATURE_DIM,), + selected_router.tensor("router_head.0.weight").float(), + selected_router.tensor("router_head.0.bias").float(), + ) + value = F.linear( + value, + selected_router.tensor("router_head.1.weight").float(), + selected_router.tensor("router_head.1.bias").float(), + ) + value = F.gelu(value) + value = F.linear( + value, + selected_router.tensor("router_head.4.weight").float(), + selected_router.tensor("router_head.4.bias").float(), + ) + return value.argmax(dim=-1) + + +def validate_groups( + features_path: Path, + groups_root: Path, + confusion_path: Path, + selected_router: ShardedCheckpoint, +) -> dict[str, Any]: + payload = torch.load(features_path, map_location="cpu", weights_only=False) + features = payload["features"] + labels = payload["labels"].long() + rows = payload.get("rows") + if not isinstance(rows, list) or len(rows) != features.shape[0] or len(rows) != labels.shape[0]: + raise ValueError("feature rows, features, and labels must have the same length") + if tuple(features.shape) != (256, 61_440): + raise ValueError(f"expected features [256, 61440], found {tuple(features.shape)}") + prediction = router_predictions(selected_router, features) + confusion = [[0 for _ in range(5)] for _ in range(5)] + for truth, pred in zip(labels.tolist(), prediction.tolist(), strict=True): + confusion[truth][pred] += 1 + correct = sum(confusion[index][index] for index in range(5)) + if correct != 205: + raise ValueError(f"selected router agreement is {correct}/256, expected 205/256") + artifact = json.loads(confusion_path.read_text()) + if artifact.get("confusion") != confusion: + raise ValueError(f"{confusion_path}: confusion does not match recomputed selected head") + + group_hashes: dict[str, Any] = {"routed": {}, "domain_assigned": {}} + for domain_id in range(5): + expected_routed = [ + row for row, pred in zip(rows, prediction.tolist(), strict=True) if pred == domain_id + ] + expected_oracle = [ + row for row, truth in zip(rows, labels.tolist(), strict=True) if truth == domain_id + ] + routed_path = groups_root / "routed_groups" / f"{domain_id}.jsonl" + oracle_path = groups_root / "oracle_groups" / f"{domain_id}.jsonl" + observed_routed = read_jsonl(routed_path) + observed_oracle = read_jsonl(oracle_path) + if [canonical_row(row) for row in observed_routed] != [ + canonical_row(row) for row in expected_routed + ]: + raise ValueError(f"{routed_path}: rows differ from selected router predictions") + if [canonical_row(row) for row in observed_oracle] != [ + canonical_row(row) for row in expected_oracle + ]: + raise ValueError(f"{oracle_path}: rows differ from offline assignments") + for condition, path, observed in [ + ("routed", routed_path, observed_routed), + ("domain_assigned", oracle_path, observed_oracle), + ]: + digests = [row_digest(row) for row in observed] + group_hashes[condition][GROUP_DOMAINS[domain_id]] = { + "path": str(path.resolve()), + "n_prompts": len(observed), + "sha256": file_sha256(path), + "canonical_row_digest_sequence_sha256": digest_sequence(digests), + } + all_digests = sorted(row_digest(row) for row in rows) + return { + "features": { + "path": str(features_path.resolve()), + "sha256": file_sha256(features_path), + "shape": list(features.shape), + "metadata": payload.get("meta"), + }, + "selected_router_input": { + "feature": "mean-only prompt pool", + "source_feature_slice": [0, ROUTER_FEATURE_DIM], + "shape": [len(rows), ROUTER_FEATURE_DIM], + "note": ( + "The retained cache also stores max/last blocks for historical analyses; " + "the selected sidecar consumes only the leading mean block." + ), + }, + "prompt_population": { + "n_prompts": len(rows), + "canonical_row_digest_set_sha256": digest_sequence(all_digests), + }, + "selected_router_recomputed": { + "assignment_agreement": correct / len(rows), + "correct": correct, + "total": len(rows), + "confusion": confusion, + "per_domain_recall": [ + confusion[i][i] / sum(confusion[i]) for i in range(5) + ], + "confusion_artifact": str(confusion_path.resolve()), + "confusion_artifact_sha256": file_sha256(confusion_path), + }, + "groups": group_hashes, + } + + +def training_state_summary(path: Path, expert_bank: Path) -> dict[str, Any]: + state = torch.load(path, map_location="cpu", weights_only=False) + args = state["args"] + init_path = Path(args.init_draft_model_path).resolve() + if init_path != expert_bank.resolve(): + raise ValueError(f"router initializer {init_path} != selected expert bank {expert_bank}") + if not bool(args.train_router_only): + raise ValueError("selected router checkpoint was not marked train_router_only") + if int(state["epoch"]) != 0 or int(state["global_step"]) != 2341: + raise ValueError("selected router training state is not epoch_0_step_2341") + return { + "path": str(path.resolve()), + "sha256": file_sha256(path), + "epoch": int(state["epoch"]), + "global_step": int(state["global_step"]), + "initializer": str(init_path), + "train_router_only": bool(args.train_router_only), + "seed": int(args.seed), + "learning_rate": float(args.learning_rate), + "batch_size_per_gpu": int(args.batch_size), + "num_domains": int(args.num_domains), + "num_anchors": int(args.num_anchors), + } + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--expert-bank", required=True, type=Path) + parser.add_argument("--selected-router", required=True, type=Path) + parser.add_argument("--exports-root", required=True, type=Path) + parser.add_argument("--features", required=True, type=Path) + parser.add_argument("--groups-root", required=True, type=Path) + parser.add_argument("--confusion-json", required=True, type=Path) + parser.add_argument("--output", required=True, type=Path) + args = parser.parse_args() + + output = args.output.resolve() + if output.exists(): + raise FileExistsError(f"refusing to overwrite {output}") + with ExitStack() as stack: + expert_bank = ShardedCheckpoint(args.expert_bank, stack) + selected_router = ShardedCheckpoint(args.selected_router, stack) + router_validation = compare_router_backbone(expert_bank, selected_router) + result = { + "analysis": "b2_epoch5_artifact_identity", + "created_utc": datetime.now(timezone.utc).isoformat(), + "torch_version": torch.__version__, + "expert_bank": { + "path": str(expert_bank.root), + "weight_artifact_sha256": expert_bank.artifact_hashes(), + "selected_code_al": 3.7186224906, + "selection_note": "best-observed code checkpoint; evaluation set also used for selection", + }, + "selected_router": { + "path": str(selected_router.root), + "weight_artifact_sha256": selected_router.artifact_hashes(), + "training_state": training_state_summary( + selected_router.root / "training_state.pt", expert_bank.root + ), + **router_validation, + }, + "standalone_exports": { + "root": str(args.exports_root.resolve()), + "validation": compare_exports(expert_bank, args.exports_root), + }, + "assignment_set": validate_groups( + args.features.resolve(), + args.groups_root.resolve(), + args.confusion_json.resolve(), + selected_router, + ), + } + output.parent.mkdir(parents=True, exist_ok=True) + temp = output.with_suffix(output.suffix + ".tmp") + temp.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + os.replace(temp, output) + print(f"wrote {output}") + + +if __name__ == "__main__": + main() diff --git a/verification/v1/verify_b5_qwen3_4b_evidence.py b/verification/v1/verify_b5_qwen3_4b_evidence.py new file mode 100644 index 0000000000000000000000000000000000000000..0311c8aee0a35033fdfc9027270887023bfff10c --- /dev/null +++ b/verification/v1/verify_b5_qwen3_4b_evidence.py @@ -0,0 +1,101 @@ +#!/usr/bin/env python3 +"""Verify the locally frozen B5 sidecars and write an artifact manifest.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +from pathlib import Path +from typing import Any + + +def file_sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for block in iter(lambda: handle.read(1024 * 1024), b""): + digest.update(block) + return digest.hexdigest() + + +def read_jsonl(path: Path) -> list[dict[str, Any]]: + rows = [] + with path.open() as handle: + for line_no, line in enumerate(handle, 1): + row = json.loads(line) + if not isinstance(row, dict): + raise ValueError(f"{path}:{line_no}: expected JSON object") + rows.append(row) + return rows + + +def resolve_unique(root: Path, remote_path: str) -> Path: + matches = list(root.rglob(Path(remote_path).name)) + if len(matches) != 1: + raise ValueError(f"expected one local copy of {remote_path}, found {len(matches)}") + return matches[0] + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--evidence-root", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + root = args.evidence_root.resolve() + bootstrap_path = root / "b5_qwen3_4b_bootstrap.json" + bootstrap = json.loads(bootstrap_path.read_text()) + provenance = bootstrap["validated_sidecar_provenance"] + expected_counts = {"code": 89, "math": 35, "factual_qa": 47, "creative_writing": 30, "general": 55} + + validated: dict[str, Any] = {} + digest_sequences: dict[tuple[str, str], list[str]] = {} + for side in ("left", "right"): + validated[side] = {} + for domain, record in provenance[side].items(): + sidecar = resolve_unique(root, record["sidecar"]) + aggregate = resolve_unique(root, record["aggregate_result"]) + if file_sha256(sidecar) != record["sidecar_sha256"]: + raise ValueError(f"local sidecar SHA mismatch: {sidecar}") + if file_sha256(aggregate) != record["aggregate_result_sha256"]: + raise ValueError(f"local aggregate SHA mismatch: {aggregate}") + rows = read_jsonl(sidecar) + if len(rows) != expected_counts[domain]: + raise ValueError(f"{sidecar}: {len(rows)} rows != {expected_counts[domain]}") + for index, row in enumerate(rows): + if row["prompt_idx"] != index or row["run_idx"] != 0: + raise ValueError(f"{sidecar}:{index + 1}: prompt order/run mismatch") + if float(row["completion_tokens"]) < 0 or float(row["spec_verify_ct"]) <= 0: + raise ValueError(f"{sidecar}:{index + 1}: invalid token counters") + digest_sequences[(side, domain)] = [str(row["prompt_digest"]) for row in rows] + validated[side][domain] = { + "n_prompts": len(rows), + "sidecar": str(sidecar.relative_to(root)), + "sidecar_sha256": file_sha256(sidecar), + "aggregate": str(aggregate.relative_to(root)), + "aggregate_sha256": file_sha256(aggregate), + } + + for domain in expected_counts: + if digest_sequences[("left", domain)] != digest_sequences[("right", domain)]: + raise ValueError(f"{domain}: left/right prompt digests differ") + + frozen_files = {} + for path in sorted(p for p in root.rglob("*") if p.is_file() and p != args.output.resolve()): + frozen_files[str(path.relative_to(root))] = { + "size_bytes": path.stat().st_size, + "sha256": file_sha256(path), + } + payload = { + "analysis": "b5_qwen3_4b_local_evidence_verification", + "bootstrap_sha256": file_sha256(bootstrap_path), + "validated_cells": validated, + "paired_prompt_digests_identical": True, + "total_prompts_per_condition": sum(expected_counts.values()), + "frozen_files": frozen_files, + } + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + print(f"verified 10 cells and wrote {args.output}") + + +if __name__ == "__main__": + main()