Skip to content

Pull requests: jd-opensource/xllm

New pull request New
40 Open 746 Closed
40 Open 746 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: support GLM5 on NPU with PD disaggregation, mtp, and reasoning parser
#928 opened Feb 14, 2026 by sunbaosong Loading…
7
feat: support glm-5 bf16 for mlu.
#927 opened Feb 13, 2026 by a120092009 Loading…
2
refactor: extract MTPWorkerImpl from SpeculativeWorkerImpl.
#926 opened Feb 12, 2026 by Clement-Wang26 Loading…
6
feat: add torch_npu version moe kernel.
#924 opened Feb 12, 2026 by liyu119 Loading…
9
feat: adapt for CANN 8.5 and PyTorch 2.7.1 for npu device.
#923 opened Feb 12, 2026 by haimbb000 Loading…
2
[Draft] bugfix: temporary workaround of mtp draft models bugs for mlu.
#919 opened Feb 12, 2026 by a120092009 • Draft
6
refactor: using llm template for all npu models.
#915 opened Feb 11, 2026 by RobbieLeung Loading…
14
feat: support multi stream for RecWorkerImpl.
#914 opened Feb 11, 2026 by Kang-Meng Loading…
7
bugfix: fix the coredump error in chat template when assistant messages are present.
#913 opened Feb 11, 2026 by DongheJin Loading…
3
refactor: clarify multimodal input error messages in callback.
#909 opened Feb 10, 2026 by xanecdotex Loading…
3
feat: support graph mode for LLM part of VLM model on NPU device.
#908 opened Feb 10, 2026 by wly-115 • Draft
feat: optimized rejected sampler for Eagle3.
#906 opened Feb 10, 2026 by RobbieLeung Loading…
2
feat: add layers that support Qwen3 model on musa device.
#902 opened Feb 10, 2026 by FleckyFelix Loading…
18
bugfix: fix deepseekv32 toolcall bugs.
#899 opened Feb 9, 2026 by DongheJin Loading…
6
refactor: abstract and refactor some similar functions in prepare_batch.
#897 opened Feb 9, 2026 by weizhehuang0827 Loading…
3
feat: support index cache transfer in PD separate deployment scenario.
#895 opened Feb 7, 2026 by sunbaosong Loading…
4
feat: adapt for CANN 8.5 and PyTorch 2.7.1 for npu device.
#891 opened Feb 6, 2026 by haimbb000 Loading…
10
bugfix: fix mrope position tensor shape mismatch in NPU graph mode.
#885 opened Feb 4, 2026 by QwertyJack Loading…
2 tasks done
1
15
feat: support mm_data rpc transport.
#876 opened Feb 3, 2026 by wly-115 Loading…
3
feat: add VMM submitter APIs for non-blocking vmm::map/unmap.
#874 opened Feb 3, 2026 by shifengmin Loading…
48
perf: optimize qwen3 moe performance on npu device.
#865 opened Feb 3, 2026 by panxua Loading…
13
feat: dynamic and scalable multi-model support.
#861 opened Feb 2, 2026 by Clement-Wang26 Loading…
22
refactor: recfactor xllm docs arch for better readability.
#845 opened Jan 31, 2026 by XuZhang99 • Draft
feat: implement rpc interface in APIService for xllm service internal usage.
#837 opened Jan 30, 2026 by weizhehuang0827 Loading…
5
bugfix: fix MTP k>1 crash by loading embed_tokens weights
#836 opened Jan 29, 2026 by QwertyJack • Draft
3 tasks done
1
6
ProTip! What’s not been updated in a month: updated:<2026-01-15.