sglang

batch_invariant_ops

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

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

batch_overlap

Optimization: Reduce the number of D2H operations (#19424 )

2026-03-04 16:32:42 +08:00

checkpoint_engine

update pre-commit config (#18860 )

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

compilation

Piecewise Cuda Graph set default (#16331 )

2026-03-02 23:18:07 +08:00

configs

Refactor NemotronHConfig to canonical layers_block_type and add MTP block-type support (#19950 )

2026-03-06 23:22:03 -08:00

connector

Remove sync points in mamba cache + prefill cudagraph plumbing for DP (#19639 )

2026-03-02 15:03:42 +08:00

constrained

Fix grammar sync across TP ranks (#17100 )

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

debug_utils

Enhance error resilience in dump comparator (#19685 )

2026-03-02 19:08:35 +08:00

disaggregation

Fix EPD OOM by offloading precomputed_embeddings during chunked prefill (#16503 )

2026-03-08 20:10:40 -07:00

distributed

[ROCm] Use unreg path for aiter custom all-reduce during CUDA graph capture (#20155 )

2026-03-09 01:09:04 -07:00

dllm

Decouple scheduler log printing from metrics collection (#20107 )

2026-03-07 22:09:10 -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 SubscribeKvEvents RPC for KV cache event streaming (#20112 )

2026-03-08 16:00:29 -07:00

eplb

Refactor: observability code cleanup (#17862 )

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

function_call

[Tool Call] Stream DeepSeek-V3.2 function call parameters in JSON format. (#16091 )

2026-03-03 01:46:29 -08:00

grpc

fix(grpc): use context.abort() with proper status codes instead of in-band errors (#19972 )

2026-03-05 14:53:18 -08:00

hardware_backend/npu

[NPU] qwen3_vl encoder support graph

2026-03-09 10:13:35 +08:00

layers

Return intermediate Mamba states (#19716 )

2026-03-09 16:04:36 +08:00

lora

[NPU] make torch_native lora backend a little bit faster (#17228 )

2026-03-07 20:14:46 +03:00

managers

Add auto bind numa node (#15678 )

2026-03-08 23:46:09 -07:00

mem_cache

Fix missing clone in hicache (#20130 )

2026-03-08 23:21:18 +08:00

model_executor

[Core] Refactor init_memory_pool into composable resolution helpers (#20142 )

2026-03-08 21:46:27 -07:00

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

Fix qwen3.5 mtp eplb related issues (#19767 )

2026-03-09 16:05:32 +08:00

multimodal

[VLM] Replace decord with torchcodec for video decoding (#20055 )

2026-03-09 19:23:49 +08:00

multiplex

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

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

observability

Decouple scheduler log printing from metrics collection (#20107 )

2026-03-07 22:09:10 -08:00

parser

rename nemotron reasoning parser (#19865 )

2026-03-05 11:27:07 -08:00

ray

Add Ray actor support for scheduler process management (DP=1) (#17684 )

2026-03-05 13:21:23 -08:00

sampling

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

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

speculative

fix: add ModelScope cache lookup and speculative path support (#20098 )

2026-03-07 22:23:16 -08:00

tokenizer

…

utils

[Bugfix] Fix load_audio: mono before resample + use torchaudio (#20054 )

2026-03-09 19:24:20 +08:00

weight_sync

…

constants.py

…

environ.py

Add auto bind numa node (#15678 )

2026-03-08 23:46:09 -07:00

server_args_config_parser.py

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

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

server_args.py

[Feature] Add SLRU eviction policy & fix RadixCache hit_count bug (#18843 )

2026-03-08 21:30:55 +08:00