sglang

batch_invariant_ops

[Auto Sync] Update batch_invariant_ops.py (20260221) (#19098 )

2026-02-20 18:27:31 -08:00

batch_overlap

[SGL] sync patch: Remove sync points, prefill cudagraph for DP, disable cache reset in mem check (#19190 )

2026-02-27 18:11:05 -08:00

checkpoint_engine

update pre-commit config (#18860 )

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

compilation

Fix PCG MoE Error (#17739 )

2026-02-19 16:48:06 +08:00

configs

Whisper model support & /v1/audio/transcriptions endpoint & benchmark (#16983 )

2026-02-23 17:28:37 -08:00

connector

[SGL] sync patch: Remove sync points, prefill cudagraph for DP, disable cache reset in mem check (#19190 )

2026-02-27 18:11:05 -08:00

constrained

Fix grammar sync across TP ranks (#17100 )

2026-01-15 18:38:01 -08:00

debug_utils

Enhance metrics in dump comparator (#19560 )

2026-02-28 18:05:19 +08:00

disaggregation

[PD] Cleanup BootstrapServer init and ready check (#19551 )

2026-02-28 16:41:42 +08:00

distributed

refactor: extract device-to-backend mapping into get_default_distributed_backend (#19202 )

2026-02-24 23:50:26 -08:00

dllm

Refactor: observability code cleanup (#17862 )

2026-02-24 18:07:29 -08:00

elastic_ep

[4/N] (Elastic EP) Back up Expert Weights in DRAM (#17374 )

2026-02-27 15:59:13 +08:00

entrypoints

feat(grpc): add multimodal TensorData parsing for vision inference (#19535 )

2026-02-27 19:29:43 -08:00

eplb

Refactor: observability code cleanup (#17862 )

2026-02-24 18:07:29 -08:00

function_call

fix KimiK2Detector regex patterns with re.DOTALL (#19120 )

2026-02-21 22:13:08 +08:00

grpc

Refactor: observability code cleanup (#17862 )

2026-02-24 18:07:29 -08:00

hardware_backend/npu

[NPU] bugs fix for Deepseek models (#19544 )

2026-02-28 17:26:15 +08:00

layers

[NPU] bugfix: resolve modelslim load weights bug (#19472 )

2026-02-28 16:22:45 +08:00

lora

[Fix] Add lora tied lm head support (for Qwen2.5, Gemma, etc model need) (#18634 )

2026-02-19 00:34:51 +08:00

managers

[NPU] bugs fix for Deepseek models (#19544 )

2026-02-28 17:26:15 +08:00

mem_cache

[SGL] sync patch: Remove sync points, prefill cudagraph for DP, disable cache reset in mem check (#19190 )

2026-02-27 18:11:05 -08:00

model_executor

[SGL] sync patch: Remove sync points, prefill cudagraph for DP, disable cache reset in mem check (#19190 )

2026-02-27 18:11:05 -08:00

model_loader

[Quantization] Support config.json quantization_config format, fix exclude_modules matching, and fix KV cache scale loading for Nemotron (#18546 )

2026-02-21 16:14:29 +08:00

models

[feat] Support nvfp4 quantized model of Qwen3-Next (#17627 )

2026-02-27 18:28:47 -08:00

multimodal

[vlm][internVL] Support processor and embedding inputs for InternVL (#19127 )

2026-02-26 22:46:48 -08:00

multiplex

[Bugfix] fix the display error (inconsistent context) (#17699 )

2026-01-31 23:35:11 -08:00

observability

Refactor: observability code cleanup (#17862 )

2026-02-24 18:07:29 -08:00

parser

fix: fix the bug blocking completion template application (#17010 )

2026-02-24 18:50:08 +08:00

sampling

Revert "feat: reduce constrained-decoding overhead in TP" (#16845 )

2026-01-10 11:39:38 +08:00

speculative

[Spec V2] Support specV2 for mamba hybrid attention (#18808 )

2026-02-27 00:36:01 +08:00

tokenizer

[Piecewise Cuda Graph] rename, refactor and add more logging (#13675 )

2025-11-21 13:28:38 +08:00

utils

Refactor: observability code cleanup (#17862 )

2026-02-24 18:07:29 -08:00

weight_sync

[RL] Allow passing tensors of different dtypes for FlattenedTensorBucket (#13413 )

2025-11-19 17:24:48 +08:00

constants.py

Support releasing CUDA graph memory when paused (#7873 )

2025-10-28 14:40:50 +08:00

environ.py

[4/N] (Elastic EP) Back up Expert Weights in DRAM (#17374 )

2026-02-27 15:59:13 +08:00

server_args_config_parser.py

[ConfigArgumentMerger] Improve ConfigArgumentMerger compatibility with external callers (#17051 )

2026-01-16 15:32:42 +08:00

server_args.py

[Session] Gate streaming sessions with --enable-streaming-session and spec v2 guard (#19531 )

2026-02-27 18:14:55 -08:00