Skip to content

Pull requests: antirez/ds4

New pull request New
286 Open 195 Closed
286 Open 195 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: code-layer performance wins (rb16 prefill dispatch, vectorized K/Q staging in the LLT scorer, double-buffered head tiles, simd_max reducers, host fixes, zero-copy payload spans)Code layer wins
#794 opened Aug 13, 2026 by daemon2k3 Loading…
agent+server: recover when a tool-call marker is truncated by the token budget
#790 opened Aug 12, 2026 by alangiu-gif Loading…
Port visible-KV checkpoint fixes for tool turns (fixes token-mismatch reprocesses on agentic workloads)
#789 opened Aug 12, 2026 by fradav Loading…
2
speed-bench: Jetson AGX Thor (sm_110) numbers
#788 opened Aug 12, 2026 by marcodelpin Loading…
1
server: stream the answer past a stray </think> instead of holding it (#783)
#787 opened Aug 12, 2026 by Flor1an-B Loading…
1
fix: detect client disconnect on plain FIN; log cancelled jobs
#785 opened Aug 12, 2026 by nazerim Loading…
1
metal: lightning-indexer-organized DS4 indexer scorer (llt)
#782 opened Aug 11, 2026 by daemon2k3 Loading…
Add M3 Ultra (512 GB) Flash benchmark.
#779 opened Aug 11, 2026 by daemon2k3 Loading…
3
dspark: fold first-token forward into verify + bit-exact M5 Metal decode wins
#778 opened Aug 11, 2026 by polymorf Loading…
metal: fix staged B-tile tensor extents in the mm_id mpp kernel
#777 opened Aug 11, 2026 by polymorf Loading…
dspark: scope scheduler no-draft pause defaults by backend
#776 opened Aug 11, 2026 by vincenzopalazzo Loading…
Exiting ds4-server when CUDA runs into terminal errors
#775 opened Aug 11, 2026 by arodd Loading…
server: return calls to custom tools as custom_tool_call
#774 opened Aug 11, 2026 by wilyan09007 Loading…
1
dspark: cut per-cycle CUDA overheads (single-draft decode, GPU confidence probe)
#772 opened Aug 10, 2026 by vincenzopalazzo Loading…
6 tasks done
metal: gate the M3-class fusion whitelists on the pre-M5 predicate
#770 opened Aug 10, 2026 by kk1987 Loading…
1
3
kv-cache: serialize only occupied compressor rows in checkpoints
#767 opened Aug 10, 2026 by kmike Loading…
1
Optimize GB10 Q2 CUDA decode while preserving bit-exact greedy output.
#766 opened Aug 10, 2026 by ivanfioravanti Contributor Loading…
6
Route batched-session requests to a slot whose KV cache they reuse
#765 opened Aug 10, 2026 by kmike Loading…
1
server: preserve custom tool kind in Responses output
#764 opened Aug 10, 2026 by carlitose Loading…
3 tasks
cuda: accelerate long-context B1 indexer and 1M HCA
#763 opened Aug 10, 2026 by wangshang23 Loading…
1
metal: accelerate M5 Max indexed prefill
#758 opened Aug 9, 2026 by rinaldofesta Contributor Loading…
1
server: anchor dsml_attr attribute match to a name boundary
#757 opened Aug 9, 2026 by Flor1an-B Loading…
tests: measure batched-verify greedy divergence rate, not just worst gap
#756 opened Aug 9, 2026 by Flor1an-B Loading…
2
CUDA: add DGX Spark network expert/tensor parallelism
#754 opened Aug 9, 2026 by shankinson Loading…
3
download: honour HF_ENDPOINT in the curl path
#753 opened Aug 9, 2026 by gaook Loading…
ProTip! Type g i on any issue or pull request to go back to the issue listing page.