Skip to content

Pull requests: antirez/ds4

New pull request New
316 Open 208 Closed
316 Open 208 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

metal: speed up IQ2_XXS MoE prefill with half LUT and split MPP
#864 opened Aug 25, 2026 by mobailabs Loading…
4 tasks done
speed-bench: Strix Halo (ROCm) results — Minisforum MS-S1 Max
#863 opened Aug 25, 2026 by alexdelprete Loading…
Consolidated ROCm exactness and performance groundwork for Strix Halo (gfx1151)
#861 opened Aug 24, 2026 by jyatesdotdev Contributor Loading…
3
server: native Prometheus GET /metrics endpoint
#857 opened Aug 23, 2026 by jyatesdotdev Contributor Loading…
2
server: fix prompt-cache miss on whitespace-only thinking channels
#856 opened Aug 23, 2026 by maxness Loading…
Serve the llama.cpp web UI from ds4-server
#854 opened Aug 22, 2026 by MiklosPathy Loading…
4 tasks done
metal: keep TP gate waits out of compute command buffers
#852 opened Aug 22, 2026 by kqb Loading…
metal: scale default prefill chunk with prompt length
#850 opened Aug 22, 2026 by kqb Loading…
ssd streaming: router-lookahead expert prefetch, +5.2% on top of #848 (output byte identical)
#849 opened Aug 21, 2026 by lukeckprobierts Loading…
ssd streaming: packed expert file, +10.7% on DeepSeek V4 Flash (output byte identical)
#848 opened Aug 21, 2026 by lukeckprobierts Loading…
metal: M1-class decode tuning, n-gram speculation, and batch-verifier groundwork
#846 opened Aug 21, 2026 by TexanoARena Loading…
Ornith 1.5 + Qwen 3.8 27B hybrid runtime + DFlash speculative decoding on Metal
#844 opened Aug 21, 2026 by audreyt Contributor Loading…
53
engine: skip the single-instance lock for --inspect
#843 opened Aug 20, 2026 by neomantra Loading…
samplers: adds top-n-sigma sampler
#841 opened Aug 20, 2026 by daniel-j-h Loading…
Distributed speculative decoding for the pipeline split (MTP + DSpark)
#835 opened Aug 19, 2026 by davidcanar Loading…
1
dspark: stop skipping after a cold miss; decode 1-token drafts
#833 opened Aug 19, 2026 by audreyt Contributor Loading…
3
metal: canonical top-k order + exact streaming top-512 selector
#832 opened Aug 18, 2026 by adriangalilea Loading…
1
metal: register-blocked, K-register-resident indexer prefill scorer (tiled4+tiled5)
#831 opened Aug 18, 2026 by adriangalilea Loading…
metal: wide-tile indexer prefill scorer over half-packed Q/K (bit-exact, +4.7% prefill at 64k)
#830 opened Aug 18, 2026 by adriangalilea Loading…
metal: non-blocking GPU stage timestamps behind DS4_METAL_GPU_STAGE_TIMESTAMPS
#828 opened Aug 18, 2026 by adriangalilea Loading…
server: idle prefill quantum follows the engine's prefill cap
#827 opened Aug 18, 2026 by adriangalilea Loading…
Reject Q2K repack candidates whose dims truncate to uint32 kernel params
#826 opened Aug 18, 2026 by x14ngch3n Loading…
1
Bound expert-output size products in ds4quant (heap overflow)
#824 opened Aug 18, 2026 by x14ngch3n Loading…
1
perf: optimize resumed streaming prefill boundaries
#822 opened Aug 17, 2026 by Elvis-Luo Loading…
server: drop a partial trailing UTF-8 sequence from finished text
#815 opened Aug 16, 2026 by adriangalilea Loading…
ProTip! Filter pull requests by the default branch with base:main.