Skip to content

Pull requests: MooreThreads/vllm-musa

New pull request New
21 Open 101 Closed
21 Open 101 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

[Performance] Unify automatic Qwen runtime fast paths
#138 opened Jul 27, 2026 by yeahdongcn Collaborator • Draft
perf(musa): add registered TP logits gather primitive
#136 opened Jul 27, 2026 by yeahdongcn Collaborator • Draft
perf(musa): fuse Qwen logits gather cast
#135 opened Jul 27, 2026 by yeahdongcn Collaborator • Draft
perf(musa): reuse Qwen uniform decode hidden state
#134 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
perf(musa): skip no-op Qwen unit-temperature divide
#133 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
perf(musa): cache Qwen uniform decode logits indices
#132 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
perf(musa): skip empty GDN graph metadata fill
#131 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
perf(musa): reuse uniform top-k in legacy Gumbel
#130 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
perf(musa): use direct FA3 schedule for Qwen graph-size-64 decode
#129 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
perf(musa): reuse uniform Qwen top-k sampling path
#128 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
perf(musa): gather Qwen TP logits through IPC
#127 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
[MUSA] Register custom all-reduce graph inputs
#126 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
[MUSA] Speed up Qwen GDN mixed-dtype causal-conv decode
#125 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
perf(musa): partition mixed Qwen seeded sampling
#124 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
perf(musa): gated Qwen-family sampling and GDN decode
#123 opened Jul 26, 2026 by yeahdongcn Collaborator • Draft
[MUSA-60042] Bypass FA3 scheduler for batch-one Qwen decode
#121 opened Jul 24, 2026 by yeahdongcn Collaborator • Draft
[MUSA-60038] Fuse Qwen3 QK norm, RoPE and KV cache update
#120 opened Jul 24, 2026 by yeahdongcn Collaborator • Draft
MUSA: switch Ubuntu 22.04 image to system Python 3.12
#119 opened Jul 24, 2026 by yeahdongcn Collaborator • Draft
1
qwen36 moe optimizations
#118 opened Jul 23, 2026 by Joey-gvwal Collaborator Loading…
Coerce hybrid-model speculative decode off full decode CUDAGraphs on MUSA
#97 opened Jul 12, 2026 by yeahdongcn Collaborator • Draft
5 tasks done
9
Fuse Qwen3 dense-decode q-norm + RoPE into one kernel (opt-in)
#90 opened Jul 6, 2026 by yeahdongcn Collaborator • Draft
6 tasks done
8
ProTip! Type g p on any issue or pull request to go back to the pull request listing page.