Skip to content

Pull requests: antirez/ds4

New pull request New
363 Open 224 Closed
363 Open 224 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: bypass DSpark speculation once a request proves unprofitable
#965 opened Sep 3, 2026 by rinaldofesta Contributor Loading…
Over 30% faster GLM 5.3 Flash decode on Metal (M3 Ultra), bit-exact
#964 opened Sep 3, 2026 by trueimage Loading…
4
ds4-agent: fix every tool call failing on DeepSeek without --vision (+ regression test)
#963 opened Sep 3, 2026 by rinaldofesta Contributor Loading…
1
server: preserve live vision cache for Pi replay
#962 opened Sep 3, 2026 by softpudding Loading…
server: persist vision KV with exact conditioning provenance
#961 opened Sep 3, 2026 by JordiPosthumus Contributor Loading…
server: restore the prompt frontier after cancelled generation
#960 opened Sep 3, 2026 by JordiPosthumus Contributor Loading…
metal: prune top-k argsort merge rounds to top_k
#959 opened Sep 3, 2026 by emilianbold Loading…
fix(server): port eval soft/hard think-close reply budgets
#958 opened Sep 3, 2026 by paride-cmd Loading…
2 of 3 tasks
1
fix(metal): coalesce adjacent --layers model map spans
#957 opened Sep 3, 2026 by paride-cmd Loading…
1 of 4 tasks
1
docs: stateless chat client live KV warm recipe
#956 opened Sep 3, 2026 by paride-cmd Loading…
2 tasks
1
docs: fix Pi Agent thinkingLevelMap for Flash modes
#955 opened Sep 3, 2026 by paride-cmd Loading…
2 tasks
1
metal: pre-M5 decode + prefill speedups — 46 t/s decode, +4% prefill on M3 Ultra (bit-exact)
#954 opened Sep 2, 2026 by ivanfioravanti Contributor Loading…
metal: exact-fp32 batched router matmul for GLM prefill (+4–8%), MoE staging ground truth
#953 opened Sep 2, 2026 by ivanfioravanti Contributor Loading…
AProjQ4: imatrix-guided Q4_K dense attention projections
#952 opened Sep 2, 2026 by GiorgioOppo Loading…
13 tasks done
3
ds4-server: make OpenAI-compatible image input work end to end
#950 opened Sep 2, 2026 by Bendr0id Loading…
server: accept inline images on tool/function role messages (fixes #933)
#944 opened Sep 2, 2026 by Flor1an-B Loading…
1
server: retry disk KV lookup with token text
#943 opened Sep 2, 2026 by JordiPosthumus Contributor Loading…
Detach tokenizer storage from model mmap
#942 opened Sep 1, 2026 by riccardomenegazzo Loading…
server: advertise loaded vision input capability
#940 opened Sep 1, 2026 by apetersson Loading…
glm: fix [gMASK]<sop> tokenization without BOS metadata
#939 opened Sep 1, 2026 by mvid Loading…
Fix DeepSeek Vision-Exp image inference with SSD streaming on Metal
#937 opened Sep 1, 2026 by cropduster Loading…
glm: keep rope models on the indexed causal attention kernel (fixes #932)
#936 opened Sep 1, 2026 by garnetlyx Loading…
1
dir-steering: pass --think, not --think-high, to ds4
#935 opened Sep 1, 2026 by eauchs Loading…
server: accept OpenAI tool-role inline images
#934 opened Sep 1, 2026 by aj-img Loading…
1
server: separate resident sessions from active requests
#930 opened Sep 1, 2026 by JordiPosthumus Contributor Loading…
2
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.