TJian 1806d1adfc [ROCm] [DSv4] [Perf] Support DeepSeek v4 MTP (#43385) ...

Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>

2026-05-24 18:43:08 +08:00

..

configs

Add tuned triton fused_moe configs on H100 for gpt-oss (#39904)

2026-04-28 03:38:48 -07:00

experts

[ROCm] [DSv4] [Perf] Support DeepSeek v4 MTP (#43385)

2026-05-24 18:43:08 +08:00

oracle

[MoE] Migrate W4A8 CT to oracle kernel setup (#42680)

2026-05-23 13:56:01 -04:00

prepare_finalize

elastic_ep: stage/commit MoE quant method on reconfigure (#40881)

2026-05-22 18:57:26 -04:00

router

[XPU]feat: add XPU fallback for MoE topk routing and MXFP4 backend (#42951)

2026-05-22 10:22:45 +00:00

runner

Revert "[Core] Replace routing replay with device cache and async D2H pipeline" (#39917) (#42434)

2026-05-13 23:49:01 -07:00

__init__.py

[MoE] Migrate W4A8 CT to oracle kernel setup (#42680)

2026-05-23 13:56:01 -04:00

activation.py

[Model] Fix Gemma4 MoE activation mismatch (#41574)

2026-05-05 04:34:11 +00:00

all2all_utils.py

elastic_ep: stage/commit MoE quant method on reconfigure (#40881)

2026-05-22 18:57:26 -04:00

config.py

Add NVFP4 MOE support for Deepseek V4. (#42209)

2026-05-22 07:21:51 -07:00

cpu_fused_moe.py

[CPU] Add MXFP4 W4A16 MoE support (#41922)

2026-05-18 03:04:45 -07:00

deep_gemm_utils.py

Fix DeepGEMM ep_scatter output address overflow (#39213)

2026-05-05 18:56:56 +00:00

eep_reconfigure.py

elastic_ep: stage/commit MoE quant method on reconfigure (#40881)

2026-05-22 18:57:26 -04:00

expert_map_manager.py

[MoE Refactor] Move expert map related code into ExpertMapManager class (#41046)

2026-05-12 09:18:27 -04:00

fused_moe_method_base.py

[MoE Refactor] Introduce RoutedExperts alias for FusedMoE and don't store SharedExperts in MK (#40735)

2026-05-12 13:37:44 -04:00

fused_moe_modular_method.py

elastic_ep: stage/commit MoE quant method on reconfigure (#40881)

2026-05-22 18:57:26 -04:00

fused_moe.py

[MoE] Move various experts classes to fused_moe/experts/ (#41979)

2026-05-11 07:54:33 +08:00

layer.py

Add NVFP4 MOE support for Deepseek V4. (#42209)

2026-05-22 07:21:51 -07:00

modular_kernel.py

elastic_ep: stage/commit MoE quant method on reconfigure (#40881)

2026-05-22 18:57:26 -04:00

moe_align_block_size.py

[Kernel][MoE] optimize moe_align_block_size (#29642)

2025-12-07 01:58:47 -08:00

moe_fused_mul_sum.py

[Quantization] add humming quantization kernel (#34556)

2026-04-24 09:29:44 -04:00

moe_permute_unpermute.py

[Refactor] Remove align block size logic in moe_permute (#33449)

2026-02-06 10:57:06 -08:00

routed_experts_capturer.py

[RFC] Replace shared-memory routed experts with ModelRunnerOutput transfer and HTTP support (#39568)

2026-05-14 14:12:30 +00:00

topk_weight_and_reduce.py

Eliminate redundant MoE buffer copies in AITER fused experts (without dependency on AITER changes) (#41713)

2026-05-07 03:46:40 -07:00

unquantized_fused_moe_method.py

[MoE Refactor] Introduce RoutedExperts alias for FusedMoE and don't store SharedExperts in MK (#40735)

2026-05-12 13:37:44 -04:00

utils.py

[Kernel] Pack topk id/weights triton kernel (#42527)

2026-05-18 03:04:31 -07:00