wxiwnd 01d6ceff8f fix(nsa): guard cp split to extend batches

2026-06-26 23:53:26 +08:00

..

attention

fix(nsa): guard cp split to extend batches

2026-06-26 23:53:26 +08:00

deep_gemm_wrapper

Reuse DeepGEMM JIT cache across restarts

2026-06-11 02:55:04 +08:00

moe

B300 NVFP4: fix FlashInferFP4MoE.forward attr names to match HEAD weight-prep

2026-06-21 16:24:17 +00:00

quantization

B300 NextN fp4: review cleanups — keep else-slice, non-routed draft trtllm, weight_scale_2 probe

2026-06-21 17:48:06 +00:00

rotary_embedding

[CPU] Add mrope kernel for Qwen3-vl (#12531)

2026-03-18 22:12:48 -07:00

utils

B300 NVFP4 port P1 stage-1: advance flashinfer_trtllm NVFP4 MoE + modelopt_fp4 path to upstream HEAD

2026-06-21 14:36:33 +00:00

activation.py

[NPU] support model skywork-reward-gemma2-2-27B-v0.2 (#16947)

2026-02-11 15:34:53 +08:00

amx_utils.py

[CPU][INT4] Add INT4 kernels for CPU (#8226)

2026-01-29 22:30:13 -08:00

communicator_nsa_cp.py

B300 CP decouple: pad-aware hidden-states gather/scatter for standard a2a under shared-KV

2026-06-21 19:06:53 +00:00

communicator.py

Support Qwen3 MoE context parallel (#18233)

2026-03-22 01:27:20 -07:00

conv.py

VLM: add Conv2dLayer/Conv3dLayer to fix PyTorch 2.9.1 CuDNN Conv3d (#20282)

2026-03-15 19:17:44 +08:00

dp_attention.py

feat(deepgemm): deepgemm jit and init in model_runner

2026-04-08 06:05:06 +08:00

elementwise.py

[Auto Sync] Update elementwise.py (20260131) (#18033)

2026-02-01 07:39:37 -08:00

flashinfer_comm_fusion.py

B300 NVFP4 port P1 stage-2: C2 flashinfer_cutedsl MoE runner + C3 flashinfer comm-fusion

2026-06-21 14:50:37 +00:00

int4fp8_utils.py

[AMD][Quantization] Add int4fp8_moe online quantization on ROCm (#7392)

2026-01-14 01:44:40 -08:00

layernorm.py

Support Qwen3 MoE context parallel (#18233)

2026-03-22 01:27:20 -07:00

linear.py

Add SGLang CUDA crash API logging inspired by FlashInfer (#20910)

2026-03-22 16:39:40 +08:00

logits_processor.py

Reduce prefill EAGLE memory pressure under CP shared KV

2026-05-13 22:29:18 +08:00

model_parallel.py

[Refactor] Rename CustomOp -> MultiPlatformOp (#16175)

2025-12-31 01:16:32 +08:00

modelopt_utils.py

Enable native ModelOpt quantization support (1/3) (#7149)

2025-10-06 13:24:15 -07:00

multimodal.py

[VLM] Improving multimodal tensor hash kernel (#9008)

2025-08-14 13:45:55 -07:00

n_gram_embedding.py

[1/n jit_kernel restruct] unify cache usage and clean up naming in ngram_embedding (#20244)

2026-03-10 15:53:43 +08:00

parameter.py

[quant] fix fp32 downcasting (#19844)

2026-03-05 17:54:59 +08:00

pooler.py

Support piecewise CUDA graph for embedding models (#13852)

2025-11-26 15:29:46 +08:00

radix_attention.py

update pre-commit config (#18860)

2026-02-16 00:18:31 +08:00

radix_linear_attention.py

[PCG] support piecewise cuda graph for kimi-linear model (#18849)

2026-02-17 23:31:12 +08:00

rocm_linear_utils.py

Aiter fp8 kv cache (#13147)

2025-12-08 16:39:53 -08:00

sampler.py

use flashinfer.sampling (#18696)

2026-02-26 10:02:38 +08:00

sparse_pooler.py

Implement BGE-M3 Sparse Embeddings in SGLang (#10869)

2025-10-22 13:46:16 -07:00

torchao_utils.py

Lazy import torchao (#17626)

2026-01-22 22:04:51 -08:00

vocab_parallel_embedding.py

Make sure we always disable symm memory without dp padding (#18129)

2026-02-04 19:58:28 +08:00