sglang

arg_groups

Check the topology identities where the layout is written, and build at the published widths (#40340 )

2026-09-21 12:22:59 -07:00

batch_invariant_ops

Add Ling-3.0-flash-VL model support (#38526 )

2026-09-17 00:19:26 +08:00

batch_overlap

[GLM-5.3-Flash] Reduce KPool planning synchronization and overlap indexer preparation (#39695 )

2026-09-19 23:51:17 -07:00

beam_search

[CI][RFC] Replace black-jupyter with ruff-format (#37210 )

2026-09-02 19:46:08 -07:00

checkpoint_engine

[XPU] Make checkpoint_engine worker device-agnostic (#32382 )

2026-09-11 09:56:39 +08:00

compilation

[SRT] Clean up no-op compiler pass, dead helpers, and migration tests (#39295 )

2026-09-16 12:04:21 +08:00

configs

[KDA] Enable ReplaySSM for GLM-5.3 Flash (#40517 )

2026-09-21 10:34:08 -07:00

connector

[NPU] Support nccl backend for --remote-instance-weight-loader (#39413 )

2026-09-17 17:28:15 +08:00

constrained

feat: use XGrammar V4.1 DSML parameter constraints (#39026 )

2026-09-21 12:12:28 -07:00

debug_utils

Give the attention-DP width and rank one home (#40067 )

2026-09-18 17:34:33 -07:00

disaggregation

Check the topology identities where the layout is written, and build at the published widths (#40340 )

2026-09-21 12:22:59 -07:00

distributed

A runner and the objects it builds freeze the placement they describe (#40341 )

2026-09-21 12:24:17 -07:00

dllm

[CI][RFC] Replace black-jupyter with ruff-format (#37210 )

2026-09-02 19:46:08 -07:00

elastic_ep

Name the two widths of the WORLD group (#40070 )

2026-09-18 17:47:02 -07:00

entrypoints

[Observability] Expose python/rust frontend identity in /server_info (#39993 )

2026-09-21 23:32:52 +08:00

eplb

[CI][RFC] Replace black-jupyter with ruff-format (#37210 )

2026-09-02 19:46:08 -07:00

function_call

feat: use XGrammar V4.1 DSML parameter constraints (#39026 )

2026-09-21 12:12:28 -07:00

hardware_backend

[npu]add chunk gdn kernel and unify ssm state layout for ascend gdn backend (#36187 )

2026-09-21 09:51:19 +08:00

kv_canary

Remove swa and mamba radix cache (#40313 )

2026-09-20 16:16:27 +08:00

layers

Check the topology identities where the layout is written, and build at the published widths (#40340 )

2026-09-21 12:22:59 -07:00

lora

[Refactor] Deduplicate kernel helpers and remove unused code (#40197 )

2026-09-19 09:27:30 +08:00

managers

A runner and the objects it builds freeze the placement they describe (#40341 )

2026-09-21 12:24:17 -07:00

mem_cache

A runner and the objects it builds freeze the placement they describe (#40341 )

2026-09-21 12:24:17 -07:00

model_executor

A runner and the objects it builds freeze the placement they describe (#40341 )

2026-09-21 12:24:17 -07:00

model_loader

Fix: post-load staging regression breaks offload meta/sharded_gpu modes (#38779 )

2026-09-21 11:19:07 +08:00

models

Check the topology identities where the layout is written, and build at the published widths (#40340 )

2026-09-21 12:22:59 -07:00

multimodal

[Model] Serve DeepSeek-OCR-2 with its official 768px local-crop geometry (#38996 )

2026-09-20 14:17:21 +08:00

multiplex

[Metrics] Propagate idle gaps across all scheduler loops (#40004 )

2026-09-18 22:42:00 -07:00

observability

State the draft's whole topology in its scope, and read the rest from the context (#40339 )

2026-09-21 12:19:38 -07:00

parser

[Fix] Add gigachat35 to the tool-call and reasoning parser name lists (#40554 )

2026-09-21 16:34:43 +08:00

platforms

Add out-of-tree DFlash extension points (#38740 )

2026-09-20 14:54:46 +08:00

plugins

[CI][RFC] Replace black-jupyter with ruff-format (#37210 )

2026-09-02 19:46:08 -07:00

ray

Record a process's placement at publish, not at group build (#40071 )

2026-09-18 23:52:37 -07:00

rust_extensions

[Rust TreeCore] Harden runtime and CI parity (#37303 )

2026-09-09 10:40:22 +08:00

rust_server

Record a process's placement at publish, not at group build (#40071 )

2026-09-18 23:52:37 -07:00

sampling

Use pinned memory for asynchronous sampling metadata transfers (#39777 )

2026-09-20 15:31:31 -07:00

session

[HiCache][Perf] fix: batch HiCache D2H submits per step for hybrid pools (#39050 )

2026-09-17 17:53:58 +08:00

speculative

State the draft's whole topology in its scope, and read the rest from the context (#40339 )

2026-09-21 12:19:38 -07:00

state_capturer

Use a shared byte budget for unified hybrid-SWA memory (#36729 )

2026-09-15 15:27:00 -07:00

tokenizer

[CI][RFC] Replace black-jupyter with ruff-format (#37210 )

2026-09-02 19:46:08 -07:00

utils

A runner and the objects it builds freeze the placement they describe (#40341 )

2026-09-21 12:24:17 -07:00

weight_cache

Check the topology identities where the layout is written, and build at the published widths (#40340 )

2026-09-21 12:22:59 -07:00

weight_sync

Introduce WeightUpdater and WeightExporter components (#31148 )

2026-07-14 15:53:33 +08:00

constants.py

Refactor server_args constants and layout (#36676 )

2026-08-27 12:16:16 -07:00

environ.py

[HiCache] TMA-staged host<->device KV transfer kernel (sm_90+) (#40278 )

2026-09-21 10:38:23 -07:00

runtime_context.py

Check the topology identities where the layout is written, and build at the published widths (#40340 )

2026-09-21 12:22:59 -07:00

server_args.py

[Perf] Fork-safe import: no CUDA context at import time, lighter argument parsing (#40201 )

2026-09-21 10:48:35 +08:00