github-featured / karpathy--llm.c

分支列表

master

f1e2ace651 · Merge pull request #801 from ngc92/ngc92/fix-test · 最后更新于 2025-05-10 23:24:10 +00:00

feature/grad_accum f133ee2294 · Merge pull request #342 from ngc92/feature/grad_accum · 最后更新于 2024-05-03 15:37:42 +00:00 frostbyte_neo	1023 3		ZIP TAR.GZ
feature/bias_backward2 776541dac7 · v1 of the new matmul backward bias kernel · 最后更新于 2024-05-02 23:32:20 +00:00 frostbyte_neo	943 0	已包含	ZIP TAR.GZ
feature/fp32_weight_master_copy c177c2694c · make nice and print · 最后更新于 2024-05-01 23:06:23 +00:00 frostbyte_neo	986 0	已包含	ZIP TAR.GZ
feature/cudnn 7d0042d574 · make training data configurable and also dont move the engire dataset to device that is not a good idea · 最后更新于 2024-05-01 18:15:10 +00:00 frostbyte_neo	1002 0	已包含	ZIP TAR.GZ
feature/packed128 af2bc47d22 · remove forceinline. blame @arund42 if this ever does not get inlined by the compiler · 最后更新于 2024-04-29 23:10:51 +00:00 frostbyte_neo	1052 0	已包含	ZIP TAR.GZ
separate_tokenizer 1443edc746 · Merge branch 'master' into separate_tokenizer · 最后更新于 2024-04-29 20:38:58 +00:00 frostbyte_neo	1067 0	已包含	ZIP TAR.GZ
ademeure-layernormbf16-3 0e4506ac30 · unify these two · 最后更新于 2024-04-29 19:03:49 +00:00 frostbyte_neo	1084 0	已包含	ZIP TAR.GZ
feature/load_bf16 9d70d9a1b2 · profile and test only use bf16. but the train script can be run with fp32 or bf16 or fp16. fp16 will error, though · 最后更新于 2024-04-28 16:17:22 +00:00 frostbyte_neo	1123 0	已包含	ZIP TAR.GZ
feature/fix_int_overflow 9c2209254b · also improve situation in the c code · 最后更新于 2024-04-26 22:27:22 +00:00 frostbyte_neo	1138 0	已包含	ZIP TAR.GZ
fix/illegalaccess b9ad95e74c · attempt to fix, but it's not working huh · 最后更新于 2024-04-14 04:40:35 +00:00 frostbyte_neo	1415 1		ZIP TAR.GZ
feature/cublaslt 44cd50cb25 · fix all the handles and things · 最后更新于 2024-04-12 22:18:18 +00:00 frostbyte_neo	1463 4		ZIP TAR.GZ

1 2 3