Skip to content

Pull requests: Luce-Org/lucebox

New pull request New
54 Open 402 Closed
54 Open 402 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

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
fix(hip): release Q8 memo before pool teardown bug Something isn't working ds4flash
#522 opened Jul 14, 2026 by dusterbloom Collaborator • Draft
perf(ggml-cuda): skip shared reduction and barrier for single-wave mul_mat_vec_q
#521 opened Jul 13, 2026 by cheese-cakee Contributor Loading…
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
perf(qwen35): add opt-in F32 rollback checkpoints for single-GPU inference
#515 opened Jul 12, 2026 by Amidwestnoob Loading…
6
feat(ds4) Restore DS4 HIP backend safely ds4flash
#514 opened Jul 12, 2026 by howard0su Contributor • Draft
11
perf(ggml/hip): add batched ROCmFPX MMQ ds4flash
#513 opened Jul 12, 2026 by weicj Collaborator • Draft
fix(ds4): correct layer-split HC handoff, per-device state, and greedy decode in local path 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 Loading…
16
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 • Draft
fix/feat(ds4): ROCmFPX byte-safe decode + fused-decode path + ds4 AR optimizations ds4flash
#503 opened Jul 8, 2026 by davide221 Contributor Loading…
10
feat(ggml): vendor ROCmFPX quant support for DeepSeek V4 Flash ds4flash
#502 opened Jul 8, 2026 by davide221 Contributor Loading…
18
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 runtime + optimizations ds4flash
#496 opened Jul 7, 2026 by davide221 Contributor • Draft
2
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
feat(dflash): add direct DeepSeek4 pure-UMA runtime path
#489 opened Jul 5, 2026 by weicj Collaborator Loading…
16
feat(ipc): optional GPU isolation for target-shard daemon (same-backend splits)
#483 opened Jul 3, 2026 by davide221 Contributor • Draft
4 tasks
feat(server): add MoE expert-split multi-target foundation
#480 opened Jul 2, 2026 by weicj Collaborator • Draft
fix(laguna) LagunaBackend ABI mismatch from KvFlashPager
#477 opened Jul 1, 2026 by davide221 Contributor • Draft
ProTip! Add no:assignee to see everything that’s not assigned.