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 260145 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 transformer/injection 分支比较 5 提交 »17c53ecb18 resolve formatting issuee436ee99df add module_inject directorya0816018aa moving the inject part to module_inject directory773aa21402 fixing the parameter data-copyf2e74517f5 ds_transformer injection code
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 transformer/large-seq-support 分支a0bd66c605 support higher sequence length in softmax backward (>2048 & <8192)
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 loadams/auto-stage3-prefetch-bucket-size 分支9c3579ff75 Test fixes on auto stage3-prefetch-bucket-size config
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 transformer-kernel/support-arbitrary-hidden 分支比较 5 提交 »9d2b71a54a support large hidden dimensions (4K and 8K)f850cc35bd supporting arbitrary hidden dimension for transformer kernelse2f204a33f add adamW and deepspeed-adamw mode283da0020c supporting cpu-adam optimizer for zero-offload on deepspeed side6faa75b1dd add adamW to CPU-ADAM implementation
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 loadams/amd-updates 分支比较 4 提交 »35aa026cb6 Move functions around29c0793330 Formatting fixesefb04f0c1b Rename AMD yaml8afc5b40f0 Update ROCm version used in AMD workflow
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 transformer/fix-layer-norm 分支比较 2 提交 »135ffe30d6 fix some issue with saving the variance for normalize forward kernelaf34cc4757 Merge branch 'master' into transformer/fix-layer-norm
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 loadams/amd-pre-compile 分支0e88c5c050 Merge branch 'master' into loadams/amd-pre-compile
- frostbyte_neo 推送了仓库 github-featured/deepspeedai--deepspeed 的 loadams/amd-mi200-tests 分支0b94abada0 Merge branch 'master' into loadams/amd-mi200-tests
