[Dependency] Upgrade to Torch 2.11.0#21247

Merged

Kangyan-Zhou merged 30 commits into

mainfrom

brayden/torch-211

May 2, 2026

Commits on May 2, 2026

more

b8zhong
authored and
Kangyan-Zhou
committed
ci: reuse sgl-kernel-build-wheels artifacts for jit-kernel tests

Kangyan-Zhou
and
claude
committed
Revert inline sgl-kernel build in execute-notebook.yml

Kangyan-Zhou
and
claude
committed
[Dependency] Bump torchcodec to 0.11.1 for torch 2.11 ABI compat

Kangyan-Zhou
and
claude
committed
[CI] Temp: force SKIP_STAGE_HEALTH_CHECK

Kangyan-Zhou
and
claude
committed
remove unnecessary reinstall for cudnn/nvshmem

Fridge003
authored and
Kangyan-Zhou
committed
update pyproject

Fridge003
authored and
Kangyan-Zhou
committed
fix: disable shared and routed expert overlap under torch.compile (not piecewise cuda graph)

b8zhong
authored and
Kangyan-Zhou
committed
fix: scale dual-stream MoE torch.compile guard by num_draft_tokens

Kangyan-Zhou
and
claude
committed
[CI] Uninstall stale cu12-suffixed packages on non-cu12 builds

authored and
Kangyan-Zhou
committed
[CI] Force-rebuild deep_ep for torch-211 PR via FORCE_REBUILD_DEEPEP

authored and
Kangyan-Zhou
committed
Revert "[CI] Uninstall stale cu12-suffixed packages on non-cu12 builds"

Fridge003
authored and
Kangyan-Zhou
committed
little update pr-test.yml

Fridge003
authored and
Kangyan-Zhou
committed
rm unnecessary h20 test

Fridge003
authored and
Kangyan-Zhou
committed
bump sglang-kernel to 0.4.2

Fridge003
authored and
Kangyan-Zhou
committed
Upload multimodal gen outputs as artifacts

mickqian
authored and
Kangyan-Zhou
committed
lint

mickqian
authored and
Kangyan-Zhou
committed
lint

mickqian
authored and
Kangyan-Zhou
committed
ci: install sgl-kernel from CU_VERSION-matched sglang wheel index

Kangyan-Zhou
and
claude
committed
ci: re-enable FORCE_REBUILD_DEEPEP for torch-211 PR

Kangyan-Zhou
and
claude
committed
ci: drop stale --auto-partition-id from stage-c-test-8-gpu-h20

Kangyan-Zhou
and
claude
committed
ci: drop SKIP_STAGE_HEALTH_CHECK temp override

Kangyan-Zhou
and
claude
committed
revert stage health check
Kangyan-Zhou
committed
Update diffusion CI thresholds for torch 2.11
mickqian
committed
Loosen qwen image edit 2511 threshold
mickqian
committed
Loosen image t2i consistency thresholds
mickqian
committed
Update 2-gpu diffusion CI thresholds
mickqian
committed
upd
mickqian
committed
Fix CLIP tokenizer RobertaProcessing fallback
mickqian
committed
Fix CLIP processor RobertaProcessing fallback
mickqian
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Dependency] Upgrade to Torch 2.11.0#21247

[Dependency] Upgrade to Torch 2.11.0#21247
Kangyan-Zhou merged 30 commits into
mainfrom
brayden/torch-211

Commits on May 2, 2026

more

ci: reuse sgl-kernel-build-wheels artifacts for jit-kernel tests

Revert inline sgl-kernel build in execute-notebook.yml

[Dependency] Bump torchcodec to 0.11.1 for torch 2.11 ABI compat

[CI] Temp: force SKIP_STAGE_HEALTH_CHECK

remove unnecessary reinstall for cudnn/nvshmem

update pyproject

fix: disable shared and routed expert overlap under torch.compile (not piecewise cuda graph)

fix: scale dual-stream MoE torch.compile guard by num_draft_tokens

[CI] Uninstall stale cu12-suffixed packages on non-cu12 builds

[CI] Force-rebuild deep_ep for torch-211 PR via FORCE_REBUILD_DEEPEP

Revert "[CI] Uninstall stale cu12-suffixed packages on non-cu12 builds"

little update pr-test.yml

rm unnecessary h20 test

bump sglang-kernel to 0.4.2

Upload multimodal gen outputs as artifacts

lint

lint

ci: install sgl-kernel from CU_VERSION-matched sglang wheel index

ci: re-enable FORCE_REBUILD_DEEPEP for torch-211 PR

ci: drop stale --auto-partition-id from stage-c-test-8-gpu-h20

ci: drop SKIP_STAGE_HEALTH_CHECK temp override

revert stage health check

Update diffusion CI thresholds for torch 2.11

Loosen qwen image edit 2511 threshold

Loosen image t2i consistency thresholds

Update 2-gpu diffusion CI thresholds

upd

Fix CLIP tokenizer RobertaProcessing fallback

Fix CLIP processor RobertaProcessing fallback