3 people 9a161307f5 [torch.compile][ROCm][V1] Enable attention output FP8 fusion for V1 attention backends (#19767) ...

Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Co-authored-by: Luka Govedič <lgovedic@redhat.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>

2025-09-10 13:59:55 -07:00

..

__init__.py

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

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

activation_quant_fusion.py

[CI] Fix broken compile tests due to unsupported SiluMul+Nvfp4Quant fusion (#23973)

2025-08-30 00:14:43 -07:00

backends.py

[torch.compile][ROCm][V1] Enable attention output FP8 fusion for V1 attention backends (#19767)

2025-09-10 13:59:55 -07:00

base_static_graph.py

Remove graph_pool as member of VllmBackend and argument to CUDAGraphWrapper (#23385)

2025-08-25 19:34:15 -07:00

collective_fusion.py

[Doc]: fix typos in Python comments (#24115)

2025-09-02 21:14:07 -07:00

compiler_interface.py

Add test case for compiling multiple graphs (#21044)

2025-07-23 11:00:47 -07:00

counter.py

[Bugfix] VLLM_V1 supports passing other compilation levels (#19340)

2025-07-29 09:35:58 -04:00

cuda_graph.py

Remove graph_pool as member of VllmBackend and argument to CUDAGraphWrapper (#23385)

2025-08-25 19:34:15 -07:00

cuda_piecewise_backend.py

[Core] Allow full cudagraph with separate attention routines and orthogonal to compilation, add support for FA2 and FlashInfer (#20059)

2025-08-15 10:01:39 -04:00

decorators.py

[torch.compile] Support conditional torch.compile per module (#22269)

2025-08-20 16:52:59 +00:00

fix_functionalization.py

[BUGFIX ] fix undefined silu_and_mul_nvfp4_quant (#23929)

2025-08-29 09:36:39 -07:00

fusion_attn.py

[torch.compile][ROCm][V1] Enable attention output FP8 fusion for V1 attention backends (#19767)

2025-09-10 13:59:55 -07:00

fusion.py

[Bugfix] Add fake mode around passes (#23349)

2025-08-28 11:25:56 -04:00

fx_utils.py

[torch.compile][ROCm] Fuse quantization onto attention using a torch.compile pass (#16756)

2025-06-12 08:31:04 -07:00

inductor_pass.py

[Bugfix] Add fake mode around passes (#23349)

2025-08-28 11:25:56 -04:00

monitor.py

[Doc]: fix typos in Python comments (#24001)

2025-08-31 08:21:59 +00:00

multi_output_match.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

noop_elimination.py

[torch.compile][ROCm] Fuse quantization onto attention using a torch.compile pass (#16756)

2025-06-12 08:31:04 -07:00

pass_manager.py

[Platform] import activation_quant_fusion for CUDA only (#23882)

2025-08-28 22:54:16 -07:00

sequence_parallelism.py

[Bugfix] Add fake mode around passes (#23349)

2025-08-28 11:25:56 -04:00

torch25_custom_graph_pass.py

[Misc] Add SPDX-FileCopyrightText (#19100)

2025-06-03 11:20:17 -07:00

vllm_inductor_pass.py

[Bugfix] Upgrade depyf to 0.19 and streamline custom pass logging (#20777)

2025-07-11 00:15:22 -07:00

wrapper.py

[Core] Allow full cudagraph with separate attention routines and orthogonal to compilation, add support for FA2 and FlashInfer (#20059)

2025-08-15 10:01:39 -04:00