Nick HillandAndreas Karatzas cc5dd0a857 [MRV1] Scope breakable cudagraphs to the piecewise path only (#56312) ...

Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>

2026-09-11 19:55:34 +00:00

..

passes

[Kernel] Enable optimized FlashInfer add-RMSNorm NVFP4 fusion (#51925)

2026-09-08 17:20:32 -07:00

__init__.py

[torch.compile] avoid Dynamo guard evaluation overhead (#7898)

2024-08-28 16:10:12 -07:00

backends.py

Revert compile-cache device index regression on CPU (#53304)

2026-08-21 16:08:13 -04:00

base_static_graph.py

Update Optional[x] -> x | None and Union[x, y] to x | y (#26633)

2025-10-12 09:51:31 -07:00

breakable_cudagraph.py

[MRV1] Scope breakable cudagraphs to the piecewise path only (#56312)

2026-09-11 19:55:34 +00:00

caching.py

[ROCm]: Bump torch 2.12, triton 3.7, torchaudio, torchvision (#50607)

2026-08-04 20:14:11 -05:00

codegen.py

[Bugfix] Fix codegen for unqualified names (#40726)

2026-05-06 01:11:37 -07:00

compiler_interface.py

[Core] Add VLLM_GPU_SYNC_CHECK env var (#44800)

2026-06-26 08:24:33 -07:00

counter.py

[compile] aot_compile should respect VLLM_DISABLE_COMPILE_CACHE (#36358)

2026-03-11 03:12:03 -07:00

cuda_graph.py

[MRV1] Scope breakable cudagraphs to the piecewise path only (#56312)

2026-09-11 19:55:34 +00:00

decorators.py

[Model Runner V2] Use MRV2 for all models by default (#53183)

2026-08-27 11:59:34 -07:00

monitor.py

[BE] Fix compile time message to be consistent (use monitoring) (#40641)

2026-04-23 00:12:08 +00:00

partition_rules.py

[Misc][BE] Type coverage for vllm/compilation [1/3] (#31554)

2026-01-06 20:37:51 -05:00

piecewise_backend.py

[Core] Label torch trace logging overhead with dynamo_timed (#39329)

2026-04-20 17:31:03 +00:00

wrapper.py

[Bugfix][V1] Clean up compiled-model bytecode hooks on VllmRunner exit (#45195)

2026-06-16 20:31:17 -07:00