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

..

__init__.py

[MoE Refactor] Create MK for TRTLLM Kernels (#32564)

2026-03-03 10:39:50 -08:00

aiter_mxfp4_w4a8_moe.py

[ROCm][Quantization][2/N] Refactor quark_moe w4a8 w/ oracle (#39136)

2026-05-04 19:50:38 -07:00

batched_deep_gemm_moe.py

[Refactor] Clean up log once scope="local" (#40540)

2026-04-22 16:42:43 -04:00

cpu_moe.py

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

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

cutlass_moe.py

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

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

deep_gemm_moe.py

[Bugfix] Add swiglu limits to deepgemm fp8 methods (#41986)

2026-05-14 15:43:13 -07:00

fallback.py

[MoE Refactor] Move remaining experts classes to experts directory (#42334)

2026-05-12 09:19:46 -04:00

flashinfer_b12x_moe.py

Integrate flashinfer b12x MoE and FP4 GEMM kernels for SM120/121 (#40082)

2026-05-20 17:21:11 +00:00

flashinfer_cutedsl_batched_moe.py

[Quantization] Add FlashInfer CuteDSL batched experts backend for NVFP4 MoE (#38251)

2026-04-06 11:57:53 -07:00

flashinfer_cutedsl_moe.py

[Perf] Re-enable flashinfer autotune by default and cleanup (#42857)

2026-05-18 09:12:27 -07:00

flashinfer_cutlass_moe.py

[Bugfix] Fix DSV4 Base model swiglu limit issue in FP8 path (#42855)

2026-05-21 19:43:01 -07:00

fused_batched_moe.py

[Bugfix] Fix DSV4 Base model swiglu limit issue in FP8 path (#42855)

2026-05-21 19:43:01 -07:00

fused_humming_moe.py

[Bugfix] fix swiglu limit issue for humming backend + deepseek v4 (#42541)

2026-05-18 17:32:26 +00:00

gpt_oss_triton_kernels_moe.py

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

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

lora_context.py

[MoE Refactor] Move remaining experts classes to experts directory (#42334)

2026-05-12 09:19:46 -04:00

lora_experts_mixin.py

[MoE Refactor] Move remaining experts classes to experts directory (#42334)

2026-05-12 09:19:46 -04:00

marlin_moe.py

Refactor AWQ Marlin MoE onto modular WNA16 oracle (#42483)

2026-05-18 08:02:43 -07:00

nvfp4_emulation_moe.py

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

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

ocp_mx_emulation_moe.py

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

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

rocm_aiter_moe.py

[ROCm][Quantization][3/N] Refactor quark_moe w4a4 w/ oracle (#41436)

2026-05-18 13:46:13 -04:00

triton_cutlass_moe.py

[MoE Refactor] Move remaining experts classes to experts directory (#42334)

2026-05-12 09:19:46 -04:00

triton_deep_gemm_moe.py

[MoE Refactor] Move remaining experts classes to experts directory (#42334)

2026-05-12 09:19:46 -04:00

triton_moe.py

[Bugfix] Fix DSV4 Base model swiglu limit issue in FP8 path (#42855)

2026-05-21 19:43:01 -07:00

trtllm_bf16_moe.py

[MoE Refactor][Test] FusedMoE layer test (#24675)

2026-04-06 17:17:23 +00:00

trtllm_fp8_moe.py

[Perf] Update TRTLLM supported MoE routing methods (#39141)

2026-04-27 14:16:22 -04:00

trtllm_mxfp4_moe.py

[Perf] Re-enable flashinfer autotune by default and cleanup (#42857)

2026-05-18 09:12:27 -07:00

trtllm_nvfp4_moe.py

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

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

xpu_moe.py

[XPU] reudce host overhead of XPU MOE (#42915)

2026-05-23 13:09:34 +08:00