-
Notifications
You must be signed in to change notification settings - Fork 106
Pull requests: jd-opensource/xllm
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
feat: introduce USE_NPU_TORCH flag for debugging and enhance NPU support for Qwen3-Dense[4/N].
#590
opened Dec 23, 2025 by
yingxudeng
Loading…
feat: add XAttention support for Qwen3 generative recommendation.
#586
opened Dec 23, 2025 by
LMX-xin
Loading…
refactor: update causal LM implementations to inherit from LlmForCausalLMImplBase[3/N].
#583
opened Dec 22, 2025 by
yingxudeng
Loading…
bugfix: fix the missing index shape in the allocat kv cache transfer.
#574
opened Dec 19, 2025 by
Clement-Wang26
Loading…
bugfix: fix several bugs in the read/write swap blocks.
#573
opened Dec 19, 2025 by
Clement-Wang26
Loading…
feat: add onerec worker impl for rec framework[5/8].
#567
opened Dec 18, 2025 by
DragonFive
Loading…
7 of 10 tasks
refactor: rename layer and model interfaces for torch_npu preparation[1/N].
#563
opened Dec 18, 2025 by
yingxudeng
Loading…
feat: add multi-priority request scheduler slidebatching.
#554
opened Dec 16, 2025 by
weizhehuang0827
Loading…
[WIP]: feat: concurrent multi stream executor for rec_model.
#548
opened Dec 15, 2025 by
zhang-minchao
Loading…
feat: initialize flashinfer planinfo at layer-0 forward stage.
#534
opened Dec 12, 2025 by
yq33victor
Loading…
feat: add a new manual management weights loader style for deepseek.
#531
opened Dec 12, 2025 by
Clement-Wang26
Loading…
refactor: remove empty_kv_cache and global_empty_kv_cache.
#514
opened Dec 10, 2025 by
RobbieLeung
Loading…
bugfix: fix the issue of ineffective input embedding transmission.
#490
opened Dec 5, 2025 by
magicheng0816
Loading…
refactor: optimize unique token count preparation of batch input builder.
#449
opened Nov 27, 2025 by
RobbieLeung
Loading…
feat: enable torch_npu graph mode for Qwen-3 dense with TP support.
#325
opened Nov 6, 2025 by
yingxudeng
Loading…
ProTip!
Add no:assignee to see everything that’s not assigned.