Skip to content

Pull requests: Luce-Org/lucebox

New pull request New
55 Open 413 Closed
55 Open 413 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

Add native Windows (MSVC) compilation support
#537 opened Jul 17, 2026 by panzhang47 Loading…
4 of 10 tasks
8
fix(layer-split): honor sampler settings + prompt history
#536 opened Jul 17, 2026 by Graffioh Contributor • Draft
bench(ds4): compare PRs with benchmarks
#535 opened Jul 16, 2026 by Graffioh Contributor • Draft
perf(ggml): reuse ROCmFP4 weights across q=4 columns
#534 opened Jul 16, 2026 by davide221 Contributor • Draft
feat: DFlash support for tensor-parallel Qwen3.6 targets
#533 opened Jul 16, 2026 by jkyamog Contributor Loading…
9
feat: Qwen3.6 target tensor parallelism
#532 opened Jul 16, 2026 by jkyamog Contributor Loading…
7
perf(ds4): compute greedy argmax on GPU (monolithic and layer split)
#531 opened Jul 16, 2026 by Graffioh Contributor • Draft
refactor(draft): codify proposal row layout
#528 opened Jul 15, 2026 by dusterbloom Collaborator Loading…
fix(ds4): honor sampler settings in decode
#527 opened Jul 15, 2026 by dusterbloom Collaborator Loading…
3
fix(ds4): use monolithic backend for single-device launches, not only for HIP
#526 opened Jul 15, 2026 by Graffioh Contributor Loading…
perf(qwen35): accelerate Ampere prefill GDN
#524 opened Jul 14, 2026 by dusterbloom Collaborator • Draft
feat(ds4): add indexed layer-major prefill ds4flash
#520 opened Jul 13, 2026 by davide221 Contributor • Draft
perf(dflash): add MMID dispatch telemetry ds4flash
#519 opened Jul 13, 2026 by cheese-cakee Contributor Loading…
7
perf(cuda): reduce MMQ stream-k overhead
#518 opened Jul 13, 2026 by dusterbloom Collaborator Loading…
refactor(backend): typing for park and unpark operations
#517 opened Jul 13, 2026 by Graffioh Contributor Loading…
3
feat(ds4) Restore DS4 HIP backend safely ds4flash
#514 opened Jul 12, 2026 by howard0su Contributor Loading…
14
perf(ggml/hip): add batched ROCmFPX MMQ ds4flash
#513 opened Jul 12, 2026 by weicj Collaborator • Draft
fix(ds4): correct layer-split HC handoff and per-device state ds4flash
#507 opened Jul 11, 2026 by Graffioh Contributor • Draft
1
perf(dflash): split-chain fast rollback with F32 SSM checkpoints (+29% decode)
#506 opened Jul 10, 2026 by Amidwestnoob Contributor Loading…
19
feat(IPC/TT): add DeepSeek4 DSpark shared worker + TP/PP optimizations ds4flash
#505 opened Jul 10, 2026 by davide221 Contributor • Draft
perf(cuda-hip): partial-top-k + hipCUB large-N for argsort/top-k on ROCm ds4flash
#504 opened Jul 10, 2026 by geometric Bot Loading…
1
chore(deps): bump the github-actions group across 1 directory with 7 updates
#498 opened Jul 7, 2026 by dependabot Bot Loading…
1
feat(ds4): add DSpark speculative decoding (+27% on Strix Halo) ds4flash
#496 opened Jul 7, 2026 by davide221 Contributor Loading…
17
feat(server): split tools from chat so PFlash works on agent turns
#492 opened Jul 5, 2026 by davidmroth Contributor Loading…
26
fix(dflash): scale DeepSeek4 graph sizing for large chunks
#490 opened Jul 5, 2026 by weicj Collaborator • Draft
3
ProTip! no:milestone will show everything without a milestone.