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 264266 次贡献
更少更多
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 find-test-failure-diff-between-envs 分支比较 2 提交 »5c1eb3ef53 Use distinct output dirs for the different runs21021a40f9 Merge branch 'main' into find-test-failure-diff-between-envs
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 feature/#35425 分支比较 8 提交 »3d401976c8 add generated outputf2bb6f982f merge with main and fix copiesee33cf7b73 ruff was updated on mainc198b4b6bb Merge branch 'main' of github.com:huggingface/transformers into feature/#35425186e32b902 update
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 v5.2-release 分支比较 4 提交 »7d9754a051 Fix unprotected torch import (#44055)a0a3d294b6 Fix peft conversion typo (#44053)2096db0aa1 v5.2.073ecc5e314 [fix][wip] GlmMoeDsa: try implement DSA (#43912)
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 feat_pp_inference 分支比较 5 提交 »b902c4086d breaking: begin ppbbc901d530 changing model._device_mesh to _tp_device_mesh (this seems to not be used anywhere anyway)37967d99f6 begin to distribute model on 2 axis7f698a4aa6 move to another folder the parallelism11d6322c0e initialize_parallelism to allow 2D mesh
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 v5.14-release 分支比较 9 提交 »a08ace4bbd Release: v4.14.11ccf4f1594 Fix deepgemm on multiple devices (#47323)c98cc10ea7 [`FP8`] Bump kernels version (#47344)e035e45e45 Fix assisted decoding for models with EncoderDecoder cache & OlmoHybrid (#47361)0eb5e032a1 Fix sdpa prefill with position_bias (#47359)
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 feat/multi-arch-ci-docker-images 分支ba4c534738 feat: build CI docker images for linux/arm64 (Graviton)
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 v5.13-release 分支比较 7 提交 »825694c414 Remove accidentally committed arch_svg/0e36730360 Trigger PR CI on v*-release branches4626421dc6 v5.13.1bef013c4a2 Be more defensive with `remap_legacy_layer_types` for custom models (#47245)8bbbe2475c Fix custom code which doesn't know about the new linear layer type names (#47174)
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 feat/kv_cache_retention_across_convos 分支比较 8 提交 »69d0e2320f feat(serve): add memory admission param to `add_request`f76bd410cb test(serve): add `ConversationCacheManager` unittest class668c8b85ec fix(serve): only evict entries that are not in use25e20d728b refactor(serve): rename ttl class and add docs699e129e75 refactor: explicit `ttl`
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 feat/hmll 分支比较 10 提交 »983c8be98a allocate target device memory before using hmllf5a2e47e3d do not load model twice1417782d46 add nvtx debug73c4b42f50 remove unecessary param for now3b9e9ffe9c improve on the capacity guard for safetensors binding
- frostbyte_neo 推送了仓库 github-featured/huggingface--transformers 的 v5.12-release 分支比较 8 提交 »ddb849abe0 v5.12.1d81db16994 mistral common backend fix (#46667)07c294c1fc Fix `peft` lower bound (#46605)e0e7504bca Revert "fix security issue with allow all kernels"7904f30943 fix security issue with allow all kernels