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/cudnn 分支比较 10 提交 »7d0042d574 make training data configurable and also dont move the engire dataset to device that is not a good idea8aa9a9ed53 and also profile lol1ed9381fd0 fix cudnnDestroy outside of ifdef69d40ebfff add readme and search for cudd in home dir35640651b3 updates to makefile: docs, override, etc
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/delete_backward_act 分支比较 10 提交 »cbe82e783a consolidate memory by deleting the grad activation struct, move some of the last pieces of memory to the forward pass3bedc9853c Merge pull request #589 from karpathy/feature/overlap-backward-reduce43a903e1b4 revert old printf0, minor touchups and comments, add assert6515f838d9 Merge branch 'comp-and-comms' of https://github.com/ngc92/llm.c into ngc92-comp-and-comms23fc8d6192f strided all-gather
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 attempt_fix_zero1 分支比较 10 提交 »05fb642ae1 attempt fix zero 17ba3ef6bac make sure to run Adam only for the range of parameters that is available on the local deviced0c014372c add ability to export fineweb100B as wellf5d041ab0c careful with NULL and checkpoint correctly273925d10d save checkpoint files to log directory, and small readability changes
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/bias_backward2 分支比较 10 提交 »776541dac7 v1 of the new matmul backward bias kernel6ebef46f83 ugh didn't notice this tiny rebasing mistake, introduced a bug. good candidate for a CI that we can overfit a single batch in the train_gpt2.cu script and get the exact same numbers as we expect in the test_gpt2.cu file2c7960040d Merge branch 'ngc92-bias-backward'937fbd8e36 small doc fixes6e48501d60 bias backward kernel that will use all available threads
- frostbyte_neo 推送了仓库 github-featured/dmlc--dgl 的 0.8.x 分支比较 3 提交 »033363eebe update version731cd90ad3 [Doc] Fix typo in nn-tensorflow.rst (#4154)faf0fd0ac8 update version
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 feature/cublaslt 分支比较 10 提交 »44cd50cb25 fix all the handles and things5364e0b853 change main train/test code to use new cublaslt functions. inference now causes segfault though, very strange have to fix before merge8cf66fbb84 change workspace from 4mib to 32mib so we don't have to detect hopper. seems ok for nowbfd11ce5b6 use cublaslt and optionally tf32, which fuses bias and deprecates the need for add_bias kernel6396e393e3 add links to discussions
- frostbyte_neo 推送了仓库 github-featured/dmlc--dgl 的 0.7.x 分支比较 10 提交 »de0eb20968 fix link prediction (#3485)582f71a173 Fix openmp header (#3325)0d1807ecf4 update version50dd8517c8 Enable faster validation for pytorch graphsage example (#3361)46557ee8a8 Update cub for cuda 11.5 compatibility (#3468)
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 master 分支比较 10 提交 »f1e2ace651 Merge pull request #801 from ngc92/ngc92/fix-test42aa89184f adjusted error thresholds166f5e6c60 reorder checks to match actual tensor order7ecd8906af Merge pull request #766 from msaroufim/patch-1315b8d1f62 Update README.md
- frostbyte_neo 推送了仓库 github-featured/karpathy--llm.c 的 ademeure-layernormbf16-3 分支比较 10 提交 »0e4506ac30 unify these two8be7370a55 gpt2_build_from_checkpointebedadfee8 Fix train_gpt2.py to write all weights as BF16 (+merge fixes)d23bb1fd13 Merge remote-tracking branch 'karpathy/master' into layernormbf16e92ec3ffd9 Merge pull request #281 from ChrisDryden/templateCleanup
