11月12月1月2月3月4月5月6月7月8月9月10月
周一周三周五
一年内 256771 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/hellaswag 分支比较 10 提交 »8d55c4a6ac fix a bug, we have to be careful to make sure we evaluate loss at the token just before the first completion token, because that is the prediction for the first token, and its accuracy4e14b52285 report both acc and acc_norm7bd2389c1f add mmlu as well and refine both a bitaec9ce5d0a move hellaswag file to dev77b991281f add hellaswag reference eval. it looks pretty bad honestly, the examples themselves, but i'm told it is predictive... hmm
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/grad_accum3 分支比较 10 提交 »8eb3a432d9 revert all streams and synchronization events, we'll bring them back but only one at a time and very very carefully, because they cause really subtle bugs and issuesa4567ae940 delete parallels, still not fixed92fc26eba4 the nuts and bolts of gradient accumulation again, merged to master, but there is a bug and it doesn't work, debugging...2d43e5bc97 remove legacy comment222d59fa2f Merge pull request #408 from ngc92/layernorm-bw-dev-cuda
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/grad_accum 分支比较 10 提交 »f133ee2294 Merge pull request #342 from ngc92/feature/grad_accumd2bd2d6f21 fixed activation gradient resetting for backward pass662f7bf545 gradient accumulation preview, doesn't work yet but i think it is 95% of the way there or so769c911ab6 small tweaksbad76f408a Merge pull request #313 from ngc92/small-improvements
- frostbyte_neo 推送了仓库 github-featured/dmlc--dgl 的 1.1.x 分支比较 10 提交 »6e2c0f489b use cccl (#6694)7cf75dd675 change version to 1.1.3 (#6643)e75910bcb1 Release 1.1.3 (#6637)6ee7cc45da [release] 1.1.2post1 (#6249)d40a3c37e5 1.1.x candidate (#6155)
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/fp32_weight_master_copy 分支比较 10 提交 »c177c2694c make nice and print795f8b690c fixes to keep master copy in fp32 of weights optionallyfd474feec8 Merge branch 'fp32-weights' of https://github.com/ngc92/llm.cdf9f0efde6 updated letter and properly free bufferf4f7a987e7 Merge pull request #324 from eymay/init_weights_from_scratch
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/gpt3v1 分支比较 10 提交 »220125f39c ok let's allow T < max_seq_len because our tests use it. make warning very prominentfaf8d159a3 use int instead of bool528b1b0e31 fix fat fingers039904ee9f revert legacy behavior in .sh scripts for -e option64db3ded85 refactor gpt2 gpt3 descriptor to keep legacy behavior of -e, be strict everywhere, remove interpolation for now
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/finetune_llama31py 分支比较 10 提交 »0aa9175569 still wip just putting things up for comment29aacba1f5 Merge pull request #705 from gordicaleksa/refactor_cef12d1b80e Merge pull request #717 from ngc92/nvml3cefe09f7e Merge branch 'master' of github.com:karpathy/llm.ce48ebc0138 Merge branch 'mspronesti-master'
- frostbyte_neo 推送了仓库 github-featured/dmlc--dgl 的 1.0.x 分支比较 10 提交 »778c367255 [Release] bump version to 1.0.4c377ca7c9f [Performance] Use binary search in rowwise pick (#5903)56c941cd43 [Release] cherry-pick DistDGL bug fixes and bump version (#5877)d9da420575 Merge branch '1.0.x' of github.com:dmlc/dgl into 1.0.x2bdea1df91 bump version
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/fix_int_overflow 分支比较 10 提交 »9c2209254b also improve situation in the c codeb019927100 change more ints to size_t for a bit more safety, still not perfect most likely689ccb8ec8 candidate fix for int overflowad602c59ff Merge pull request #239 from dagelf/master085270618d upgrade our .py script so it can also do torchrun with many processes, as our mixed precision mpi nccl code can do now
- frostbyte_neo 推送了仓库 github-featured/dmlc--dgl 的 0.9.x 分支比较 8 提交 »d1517f4797 update version0bd6bc80cd [examples]educe memory consumption (#4558) (#4596)960092be02 [Feature] Import PyTorch's CUDA stream management (#4503)75d23de415 update conda scripts0769122b65 update version
