Skip to content

Pull requests: jd-opensource/xllm

New pull request New
52 Open 835 Closed
52 Open 835 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

feat: optimize cal_seq_exchange_index using unordered_map.
#1041 opened Mar 11, 2026 by Dragonliu2018 Loading…
bugfix: add synchronize when running multi-modal embedding model.
#1040 opened Mar 11, 2026 by wxh571001500 Loading…
feat: add embedding manager to reduce total size of embeding cache.
#1039 opened Mar 11, 2026 by RobbieLeung Loading…
4
feat: add unified request stats logs for prefill and service.
#1038 opened Mar 11, 2026 by DongheJin Loading…
3
feat: lm head uses row parallel to support more sizes of vocab.
#1037 opened Mar 11, 2026 by RobbieLeung Loading…
2
refactor: remove the outdated glm5 model type.
#1034 opened Mar 11, 2026 by yq33victor Loading…
4
bugfix: adjust acl graph sequence capacity for speculative modes.
#1030 opened Mar 10, 2026 by DongheJin Loading…
3
feat: support glm-5 w8a8 for mlu.
#1029 opened Mar 10, 2026 by phantomlei3 Loading…
6
feat: support QwenImageEditPlus pipeline with embedding infer.
#1028 opened Mar 10, 2026 by shan-chen-feng Loading…
8
feat: add mlu sequence parallel prerequisites part 2.
#1026 opened Mar 10, 2026 by phantomlei3 Loading…
16
perf: optimize FP8 GEMM performance via tile strategy tuning[3/N].
#1025 opened Mar 9, 2026 by yingxudeng Loading…
11
feat: support rolling weight loading.
#1024 opened Mar 9, 2026 by Clement-Wang26 Loading…
10
refactor: extract multi-modal input processors to processors dir.
#1022 opened Mar 9, 2026 by wly-115 Loading…
3
feat: pin xllm process to the right numa region.
#1014 opened Mar 8, 2026 by Dragonliu2018 Loading…
1
14
github: only build on PR approval, skip on raw push/open.
#1012 opened Mar 6, 2026 by Clement-Wang26 Loading…
1
bugfix: improve qwen25 tool-call parsing robustness.
#1010 opened Mar 6, 2026 by yingxudeng Loading…
3
refactor: unify rec multi round decode mode with one-stage flag.
#1000 opened Mar 5, 2026 by LMX-xin Loading…
7
bugfix: add group_id when creating the HccLProcessGroup to support multiple communication process_group.
#999 opened Mar 5, 2026 by shan-chen-feng Loading…
6
[WIP] feat: add onerec model implement[3/N]
#998 opened Mar 4, 2026 by DragonFive Loading…
3
feat: support prefix cache for multi-modal model.
#997 opened Mar 4, 2026 by wly-115 • Draft
@wly-115
5
feat: Support Qwen3-next on NPU device
#989 opened Mar 4, 2026 by JC-ut0 Loading…
14
feat: add qwen35 and qwen35-thinking reasoning parser detectors.
#985 opened Mar 3, 2026 by yingxudeng Loading…
7
feat: support qwen3.5 tool-call parser with qwen3_coder detector.
#982 opened Mar 3, 2026 by yingxudeng Loading…
3
feat: add onerec model implement[2/N]
#981 opened Mar 3, 2026 by DragonFive Loading…
18
feat: support xLLM CUDA backend integration with tilelang kernels.
#980 opened Mar 3, 2026 by Wang-1F • Draft
6
ProTip! Filter pull requests by the default branch with base:main.