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 260013 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 gma/xpu_compile_analysis 分支比较 2 提交 »dcf7af9a50 remove unneeded comments and add zero 212af27d430 Merge branch 'master' into gma/xpu_compile_analysis
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 gma/update_agents_md 分支809c56c98c docs: add test directory convention to AGENTS.md
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 gma/superoffload_nogpufp32grad 分支比较 5 提交 »e9227620d9 allow OMP_NUM_THREADS in superoffload33b9dab99a add comments for failure0d15c5f2d8 fix some bugs1978617c69 allow config grad convert on CPU018bb69f3e Provide option to not convert GPU grad to FP32 before copy to CPU
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 qanthony/comms-bench 分支比较 10 提交 »45e6b350a8 Fixed allgather mem_factorc6370044d9 Fix ds_bench, fix args, fix dtypes, fix bw_unitc3ec062ff6 No longer package ds_bench with setup.py240df2ca7b Add ds_bench to setup and READMEbabab4f7f1 Added ds_bench, abstracted out args, error handling
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 gma/opd_rollout_engine_test 分支比较 10 提交 »918b69043b fix(rollout): fix CUDA graph capture producing incorrect logits5e4e8d124d expand question list to 200+f9a73e0631 rollout: add FlashInfer kernel manager and vLLM compat shim60dd200a85 fix: AutoTP _replace_module uses full hierarchical path for pattern matchingdb8289adb4 rollout: fix hybrid engine device index bug, vllm import path, and smoke configs
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 qanthony/bigbird 分支a9c942119b Merge branch 'master' into qanthony/bigbird
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 preserve-CVDs 分支8639b3c476 preserve cuda visible devices order
