12月1月2月3月4月5月6月7月8月9月10月11月
周一周三周五
一年内 259992 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 fs-z2-fix 分支9bd26641e2 Update stage_1_and_2.py
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 olruwase/zero2_offload_rrb_divergence 分支7030346c77 Merge branch 'master' into olruwase/zero2_offload_rrb_divergence
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 fs-soft-kernel 分支比较 3 提交 »70420bf3fe fix softmax issue & curand_generator on A100ea87c94ce4 add unit test for checking new kernel functionality5537e8c85a add the softmax-dropout kernel
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 fs-82 分支26c51b466f less aggressive prefetching
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 olruwase/zero2_offload_keyerror 分支比较 3 提交 »bbc475e29a Merge branch 'olruwase/zero2_offload_keyerror' of github.com:microsoft/DeepSpeed into olruwase/zero2_offload_keyerrorb55cbf3e94 Merge branch 'master' of github.com:microsoft/DeepSpeed into olruwase/zero2_offload_keyerrorab027ee9a7 Merge branch 'master' into olruwase/zero2_offload_keyerror
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 fp6-blog 分支6e4f0a0a3b Merge branch 'master' into fp6-blog
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 olruwase/zero2_grad_accum_bug 分支比较 4 提交 »ce5d907488 Formatting fixes6a7829a163 Merge branch 'master' of github.com:microsoft/DeepSpeed into olruwase/zero2_grad_accum_buge1ff8babf8 Unit test for unbalanced gradients in ZeRO0bb0cc8061 Use zero-tensors for missing gradients to avoid size mismatch
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 flops-profiler-skip-unused-args 分支96c8f82d18 Merge branch 'master' into flops-profiler-skip-unused-args
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 olruwase/zcode_model_expert 分支310d1df9b7 Early hacks
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 flash-attention 分支比较 2 提交 »ea24636398 add flash attention4125734689 Merge branch 'master' of github.com:microsoft/DeepSpeed