-
Notifications
You must be signed in to change notification settings - Fork 87
Pull requests: jd-opensource/xllm
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
refactor: separate mlu and cuda version Qwen model implementation.
#468
opened Dec 1, 2025 by
XuZhang99
Loading…
feat: determine device type automatically except npu device.
#467
opened Dec 1, 2025 by
XuZhang99
Loading…
feat: add NPU process group initialization and management.
#456
opened Nov 29, 2025 by
yingxudeng
Loading…
feat: implement chunked prefill and prefix cache for Qwen3 MoE.
#455
opened Nov 29, 2025 by
yingxudeng
Loading…
refactor: replace decode_seq_range with batch_forward_type.
#451
opened Nov 27, 2025 by
RobbieLeung
Loading…
feat: support qwen2_5_vl/qwen3_vl/qwen3_vl_moe on mlu device.
#450
opened Nov 27, 2025 by
a120092009
Loading…
refactor: optimize unique token count preparation of batch input builder.
#449
opened Nov 27, 2025 by
RobbieLeung
Loading…
refactor: move draft input preparation of decode batch from worker to batch builder.
#448
opened Nov 27, 2025 by
RobbieLeung
Loading…
[WIP] feat: support loading model weights and forward overlap.
#441
opened Nov 26, 2025 by
Clement-Wang26
Loading…
feat: enhance Qwen3-MoE to support TP settings beyond 4.
#427
opened Nov 24, 2025 by
yingxudeng
Loading…
feat: support Qwen2-VL & GME-Qwen2-VL model on npu device.
#399
opened Nov 18, 2025 by
xanecdotex
Loading…
feat: enable torch_npu graph mode for Qwen-3 dense with TP support.
#325
opened Nov 6, 2025 by
yingxudeng
Loading…
ProTip!
Follow long discussions with comments:>50.