bnellnm 2902c34826

[Kernels] Remove BatchedTritonOrDeepGemmExperts and default fallback to Triton (#29929) ...

Signed-off-by: Bill Nell <bnell@redhat.com>
Signed-off-by: bnellnm <49004751+bnellnm@users.noreply.github.com>
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>

2025-12-03 20:49:00 +00:00

..

configs

Add Mistral Large 3 and Ministral 3 (#29757)

2025-12-02 10:29:00 +00:00

__init__.py

[Kernels] Remove BatchedTritonOrDeepGemmExperts and default fallback to Triton (#29929)

2025-12-03 20:49:00 +00:00

all2all_utils.py

[Feat][Perf] Enable deepep-low-latency with round-robin expert placement. (#28449)

2025-11-19 13:46:24 +01:00

batched_deep_gemm_moe.py

[Performance][DeepGEMM] Estimate expected_m (#28694)

2025-11-15 13:52:14 +08:00

config.py

Handle triton kernel import exception (#29062)

2025-11-22 10:07:50 +00:00

cpu_fused_moe.py

[fix][cpu] Use a SwigluOAI impl which supports interleaved gate-up wei (#29273)

2025-11-25 15:11:11 +08:00

cutlass_moe.py

…

deep_gemm_moe.py

[Performance][DP/EP] Add silu_mul_per_token_group_quant_fp8_colmajor kernel (#29470)

2025-12-03 18:04:59 +00:00

deep_gemm_utils.py

…

deepep_ht_prepare_finalize.py

…

deepep_ll_prepare_finalize.py

[MoE] CuteDSL MoE with Nvfp4 DeepEP dispatch (#27141)

2025-11-30 16:05:32 -08:00

flashinfer_cutedsl_moe.py

[MoE] CuteDSL MoE with Nvfp4 DeepEP dispatch (#27141)

2025-11-30 16:05:32 -08:00

flashinfer_cutlass_moe.py

Feature: Support Relu2 in FusedMoE fp8 cutlass path (#27261)

2025-11-16 13:39:44 -05:00

flashinfer_cutlass_prepare_finalize.py

[Kernels] Enable FlashInfer FP8 Blockscale on SM90 (for TEP DSR1) (#27134)

2025-11-14 08:02:44 -08:00

flashinfer_trtllm_moe.py

[Performance] Support FP8 flashinfer TRTLLM MOE on Qwen3 and Qwen-3next (#27492)

2025-11-10 12:34:57 -05:00

fused_batched_moe.py

…

fused_marlin_moe.py

[Kernel][Quantization] add w4a8 support for marlin kernel (#24722)

2025-11-29 07:19:33 -08:00

fused_moe_method_base.py

[MoE][Refactor] Make select_experts a non-static method (#29067)

2025-11-24 13:38:04 -05:00

fused_moe_modular_method.py

[MoE][Refactor] Make select_experts a non-static method (#29067)

2025-11-24 13:38:04 -05:00

fused_moe.py

[Log] Optimize startup log (#28948)

2025-11-21 08:46:20 -08:00

gpt_oss_triton_kernels_moe.py

[LoRA] Support FusedMoE LoRA Triton kernel for mxfp4 (#29708)

2025-11-30 10:37:25 +08:00

layer.py

[Feat] Support non-gated activations in NVFP4 modelopt path (#29004)

2025-11-30 11:02:40 -05:00

modular_kernel.py

[BugFix] Make sure to allocate worst case MoE workspace during profile run in the DP + EP case (#27426)

2025-11-21 11:41:52 -08:00

moe_align_block_size.py

…

moe_pallas.py

…

moe_permute_unpermute.py

…

moe_torch_iterative.py

…

pplx_prepare_finalize.py

…

prepare_finalize.py

[Feature] Shared Experts Overlap with FI deepgemm swap kernel, 2.2% throughput improvement and 3.6% TTFT improvement (#28879)

2025-11-20 18:41:49 -07:00

rocm_aiter_fused_moe.py

[RFC][ROCm][AITER] Keep all AITER kernels in _aiter_ops class like _custom_ops and _ipex_ops (#24490)

2025-11-10 08:20:53 -08:00

routing_simulator.py

…

shared_fused_moe.py

[Bugfix] fix IMA issue in certain cases of the moe marlin kernel (#28619)

2025-11-26 19:02:21 -08:00

topk_weight_and_reduce.py

…

triton_deep_gemm_moe.py

…

trtllm_moe.py

…

unquantized_fused_moe_method.py

[MoE][Refactor] Make select_experts a non-static method (#29067)

2025-11-24 13:38:04 -05:00

utils.py

…