Skip to content

Pull requests: jd-opensource/xllm

New pull request New
81 Open 1,405 Closed
81 Open 1,405 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

refactor: precompute per-sequence block hashes to avoid recomputation.
#1657 opened Jun 5, 2026 by Clement-Wang26 Collaborator Loading…
5 of 17 tasks
6
perf: reduce greedy decode host overhead.
#1654 opened Jun 4, 2026 by pjgao Contributor Loading…
2
feat: support CANN 9.0.0 toolkit for npu device.
#1653 opened Jun 4, 2026 by fengz72 Loading…
8 of 17 tasks
2
feat: add support for DSA topk sharing on NPU devices.
#1652 opened Jun 4, 2026 by sanlio36 Collaborator Loading…
9 of 17 tasks
3
feat: support W8A8 dynamic quantization for DiT models qwen image edit.
#1649 opened Jun 4, 2026 by ethan686 Contributor Loading…
5
bugfix: fix qwen3.5 gdn precision issue
#1643 opened Jun 3, 2026 by fengz72 Loading…
8 of 17 tasks
7
perf: support aclnnsiwglu & aclnnmatmul for Atb layers on npu device.
#1642 opened Jun 3, 2026 by shan-chen-feng Collaborator Loading…
17 tasks
3
perf: optimize DeepSeek V4 NPU decode metadata.
#1632 opened Jun 2, 2026 by DongheJin Collaborator Loading…
17 tasks
4
bugfix: apply SwiGLU clamp in MoE.
#1624 opened Jun 1, 2026 by JC-ut0 Contributor Loading…
8 of 17 tasks
4
bugfix: fix qwen3.5 mtp validate attention path
#1621 opened Jun 1, 2026 by pjgao Contributor Loading…
3
bugfix: zero-initialize default-constructed XXH3Key to avoid undefined behavior.
#1618 opened Jun 1, 2026 by yingxudeng Collaborator Loading…
17 tasks
2
feat: support DiT and VAE for flux2
#1615 opened May 31, 2026 by wang-shuibin Loading…
12 tasks
16
feat: support wan22 pipeline.
#1608 opened May 30, 2026 by ethan686 Contributor Loading…
32
feat: add mlu deepseek v4 dsa metadata support.
#1602 opened May 29, 2026 by phantomlei3 Collaborator Loading…
11 of 17 tasks
9
Decode conv1d tilelang
#1598 opened May 29, 2026 by BikingNow Loading…
5 of 17 tasks
16
Modify XLLM_OPS dependency: xllm -> custom_xllm_math
#1568 opened May 26, 2026 by ware2009 Loading…
3 of 11 tasks
2
feat: parallelize manual-loader decoder weight merge.
#1542 opened May 25, 2026 by Clement-Wang26 Collaborator Loading…
6
[WIP]perf: overlap mtp draft extend preparation
#1541 opened May 25, 2026 by pjgao Contributor Loading…
6
feat: support Cola-DLM on cuda device.
#1539 opened May 24, 2026 by Dragonliu2018 Collaborator • Draft
9
feat: support interlayer add norm and SplitRmsnormRope operation for qwen3.
#1531 opened May 22, 2026 by shan-chen-feng Collaborator Loading…
5
refactor: remove xattention one-stage decode path.
#1504 opened May 21, 2026 by LMX-xin Collaborator • Draft
2
feat: support vae parallel for qwen-image-edit-plus.
#1499 opened May 20, 2026 by shan-chen-feng Collaborator Loading…
13
feat: add TileLang chunk_gated_delta_rule_fwd_h kernel.
#1498 opened May 20, 2026 by fengz72 Loading…
5
feat: support customized multimodal preprocess configs.
#1481 opened May 19, 2026 by xanecdotex Collaborator Loading…
4
refactor: remove negative condition when choosing decode or prefill
#1475 opened May 18, 2026 by rauletorresc Contributor Loading…
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.