Skip to content

Pull requests: xLLM-AI/xllm

New pull request New
116 Open 1,895 Closed
116 Open 1,895 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

perf: optimize DeepSeek V3.2 Python execution.
#2199 opened Aug 11, 2026 by yt-zeng Collaborator Loading…
1 of 7 tasks
bugfix: fix qwen3.5 vlm smoothquant mlu vision path
#2198 opened Aug 11, 2026 by phantomlei3 Collaborator Loading…
10 of 17 tasks
refactor: extract AuxHiddenCapture for spec draft hidden states.
#2197 opened Aug 11, 2026 by ustcfy Collaborator Loading…
5 of 16 tasks
1
perf: drop multimodal mrope handling on npu decode path.
#2196 opened Aug 11, 2026 by wly-115 Collaborator Loading…
7 of 17 tasks
perf: add asynchronous host kv cache transfers.
#2195 opened Aug 11, 2026 by bupt-ycq Contributor Loading…
17 tasks
feat: support DeepSeek V4 DSpark decoding on release/v0.11.0.
#2194 opened Aug 11, 2026 by chenchuw886 Contributor Loading…
12 of 17 tasks
fix: prefer top-level config.json for backend detection.
#2192 opened Aug 11, 2026 by weizhehuang0827 Collaborator Loading…
4 tasks
bugfix: cover MTP decode graph buckets.
#2190 opened Aug 11, 2026 by phantomlei3 Collaborator Loading…
10 of 15 tasks
3
bugfix: reject prompts exceeding decode kv capacity.
#2188 opened Aug 11, 2026 by shifengmin Collaborator Loading…
9 of 17 tasks
2
perf: adapt Qwen W8A8 quantization for FlashComm.
#2187 opened Aug 11, 2026 by Cooofish Contributor Loading…
2 of 7 tasks
1
feat(sfa): override sparse_flash_attention to PR#6 contract (3 outputs / 8 attrs, rope=None supported)
#2185 opened Aug 11, 2026 by 0moyi0-2024 Loading…
bugfix: fix bug for mix usage of common / atb model namespace registr…
#2184 opened Aug 11, 2026 by shan-chen-feng Collaborator Loading…
17 tasks
feat(lora): multi-tenant HTTP hot-swap API + cross-node broadcast + dense qwen3 preload
#2181 opened Aug 10, 2026 by cchh05 Collaborator Loading…
3 tasks
feat: harden EPLB graph accounting and configuration.
#2180 opened Aug 10, 2026 by DongheJin Collaborator Loading…
17 tasks
build: support standalone MUSA builds.
#2179 opened Aug 10, 2026 by fay85 Collaborator Loading…
4 of 17 tasks
[WIP]feat: context-parallel sequence sharding for Qwen3 Python model.
#2178 opened Aug 10, 2026 by yingxudeng Collaborator • Draft
17 tasks
refactor: add MUSA runtime integration.
#2176 opened Aug 10, 2026 by fay85 Collaborator Loading…
4 of 17 tasks
bugfix: separate greedy and probabilistic draft proposals for rejection sampling.
#2175 opened Aug 10, 2026 by ustcfy Collaborator • Draft
3 of 7 tasks
feat: support DeepSeek V4 DSpark decoding.
#2172 opened Aug 9, 2026 by chenchuw886 Contributor Loading…
12 of 17 tasks
21
bugfix: fix A5 TileLang build and mRoPE correctness.
#2171 opened Aug 8, 2026 by wuyifan2017 Contributor Loading…
6 of 17 tasks
bugfix: canonicalize empty DP decode metadata for hybrid ACL graphs.
#2170 opened Aug 8, 2026 by ShareableXue Contributor Loading…
10 of 17 tasks
1
build: optimize native and tilelang incremental builds.
#2168 opened Aug 7, 2026 by jarvis666666 Loading…
4 of 17 tasks
perf: optimize GLM MTP scheduler overlap.
#2167 opened Aug 7, 2026 by sanlio36 Collaborator Loading…
17 tasks
feat: add residual-aware SJF scheduling policy for PD prefill.
#2164 opened Aug 7, 2026 by immengzi Loading…
6 of 13 tasks
bugfix: adjust MTP batch capacity for concurrency.
#2158 opened Aug 7, 2026 by Rellion-926 Contributor Loading…
7 of 17 tasks
3
ProTip! no:milestone will show everything without a milestone.