sglang

benchmark

Optimize ngram decode id computation (#24757 )

2026-06-02 17:37:34 +08:00

csrc

Feat/add w4a16 moe support to nemotron (#25655 )

2026-06-02 22:42:26 -07:00

diffusion

[diffusion][jit_kernel] perf: varlen FA fast path for USPAttention masked branch (#26318 )

2026-05-28 21:26:09 +08:00

dsv4

Add FP4 Indexer for DeepSeek V4 (#26209 )

2026-06-02 00:14:38 -07:00

include/sgl_kernel

[AMD] Dsv4/pr1 fix run time issue (#25898 )

2026-05-23 16:04:14 -07:00

kv_canary

Add token-id verification to the KV-canary (#26818 )

2026-05-31 09:58:51 +08:00

tests

Feat/add w4a16 moe support to nemotron (#25655 )

2026-06-02 22:42:26 -07:00

triton

[GDN] Optimize prefill QKV split dispatch (#26206 )

2026-06-02 16:48:31 +08:00

__init__.py

Move fa4 from sgl-kernel to jit kernel (#17353 )

2026-01-24 15:25:03 +08:00

__main__.py

[Chore] Clean up JIT compilation flags (#21022 )

2026-03-25 18:08:40 +08:00

.clang-format

[JIT Kernel][Feature] Support JIT custom all reduce (rewrite as v2) (#19880 )

2026-03-20 18:24:07 +08:00

activation.py

[MoE] Deprecate act_and_mul_triton; fold filter_expert into JIT silu/gelu_and_mul (#23707 )

2026-04-26 01:41:35 -07:00

add_constant.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

all_reduce.py

[Feature][JIT Kernel] Fused TP QK norm For Minimax (#20673 )

2026-04-13 20:29:47 +08:00

awq_dequantize.py

[Kernel Slimming] Migrate AWQ marlin repack kernel to JIT (#18949 )

2026-02-23 22:05:27 +08:00

awq_marlin_repack.py

[misc] clean up kernel API (#21325 )

2026-03-25 09:10:23 +08:00

clamp_position.py

Replace clamp_position with JIT kernel + platform dispatch (#20999 )

2026-03-21 21:26:26 -07:00

concat_mla.py

[Move sgl-kernel Kernel to JIT] Add JIT concat MLA kernels (#17889 )

2026-02-03 10:49:17 +08:00

cutedsl_gdn.py

[jit-kernel] Add CuTe DSL GDN Decode Kernel (#15631 )

2026-01-18 12:54:36 -08:00

cutedsl_kda.py

[KDA] Support CuTeDSL KDA decode kernel (#21203 )

2026-03-25 09:47:09 +08:00

fixup_zero_kv.py

Fix trtllm mla chunked-prefill zero-length bug (#22291 ) (#22688 )

2026-04-20 22:10:13 -07:00

flash_attention_v3.py

[diffusion] [AMD] feat: support online MXFP4 and fp8 quantization (#21431 )

2026-05-14 08:52:01 +08:00

flash_attention_v4.py

Kernels community fa3 (#20796 )

2026-04-07 12:48:44 -07:00

flash_attention.py

perf: eliminate attention DtoD copy by passing pre-allocated output to FA (#21985 )

2026-04-24 12:05:16 -07:00

fp8_quantize.py

[perf] prepare_prefill_qkv hook + fp8 quantize jit kernel (#25460 )

2026-05-20 14:20:49 -07:00

fused_metadata_copy.py

[Refactor] Rename NSA → DSA: user-facing aliases, file/class/import rename (#25821 )

2026-05-20 00:18:04 -07:00

fused_qknorm_rope.py

[jit_kernel] Optimize fused_qknorm_rope: deduplicate sincosf for interleave RoPE (#21654 )

2026-04-01 09:04:13 +08:00

fused_store_index_cache.py

[Refactor] Rename NSA → DSA: user-facing aliases, file/class/import rename (#25821 )

2026-05-20 00:18:04 -07:00

gptq_marlin_repack.py

[misc] clean up kernel API (#21325 )

2026-03-25 09:10:23 +08:00

gptq_marlin.py

[misc] clean up kernel API (#21325 )

2026-03-25 09:10:23 +08:00

grouped_topk.py

support Hy3 preview (#23533 )

2026-04-24 12:03:24 -07:00

hadamard.py

Support piecewise CUDA graph with NSA (#23351 )

2026-05-22 14:39:50 -07:00

hicache.py

[Hicache & JIT_kernel] Support page first layout & mla jit kernel (#18311 )

2026-03-27 08:54:36 -07:00

hisparse.py

Deepseek V4 (#23882 )

2026-05-07 18:32:21 -07:00

kvcache.py

[misc] clean up kernel API (#21325 )

2026-03-25 09:10:23 +08:00

mla_kv_pack_quantize_fp8.py

perf(mla): hybrid Triton fused cat+FP8-quantize for MLA chunked-prefill K/V (#25333 )

2026-05-15 10:51:00 -07:00

moe_align.py

[sgl-kernel] support > 1024 experts in moe_align_block_size kernel (#21610 )

2026-04-08 11:45:13 -07:00

moe_fused_gate.py

Deepseek V4 (#23882 )

2026-05-07 18:32:21 -07:00

moe_lora_align.py

[5/n] Lora support cuda graph (#21647 )

2026-04-04 00:31:46 -07:00

moe_wna16_marlin.py

Revert "[Feature] NVFP4 Marlin fallback for non-Blackwell GPUs (SM75+… (#22047 )

2026-04-03 13:12:30 -07:00

mxfp8.py

[JIT Kernel][1/2]Migrate MXFP8 Group GEMM & Quant into JIT (#23833 )

2026-04-29 22:50:09 +08:00

ngram_corpus.py

[Spec][Ngram] Clean up unused stateless batchMatch (#22487 )

2026-04-10 21:52:56 -07:00

ngram_embedding.py

Optimize ngram decode id computation (#24757 )

2026-06-02 17:37:34 +08:00

norm.py

[bugfix] Honor cast_x_before_out_mul in RMSNorm.forward_cuda residual path (#25920 )

2026-05-28 01:22:46 -07:00

nvfp4.py

[diffusion] quant: add FLUX.1-dev modelopt nvfp4 support (#22672 )

2026-04-14 15:00:59 +08:00

per_tensor_quant_fp8.py

[misc] clean up kernel API (#21325 )

2026-03-25 09:10:23 +08:00

per_token_group_quant_8bit.py

[misc] clean up kernel API (#21325 )

2026-03-25 09:10:23 +08:00

resolve_future_token_ids.py

Replace _resolve_future_token_ids with JIT kernel + platform dispatch (#20976 )

2026-03-20 01:47:03 -07:00

rmsnorm_hf.py

[Fix/Kernel] Add JIT rmsnorm_hf kernel to fix transformers backend MMLU accuracy regression (#22931 )

2026-04-23 12:00:31 +08:00

rope.py

[misc] clean up kernel API (#21325 )

2026-03-25 09:10:23 +08:00

set_mla_kv_buffer.py

perf(mla): TMA bulk-store set_mla_kv_buffer (up to 12× over baseline) (#25311 )

2026-05-14 18:23:41 -07:00

timestep_embedding.py

[misc] clean up kernel API (#21325 )

2026-03-25 09:10:23 +08:00

triton_store_cache.py

[AMD] Dsv4/pr1 fix run time issue (#25898 )

2026-05-23 16:04:14 -07:00

utils.py

Feat/add w4a16 moe support to nemotron (#25655 )

2026-06-02 22:42:26 -07:00