Skip to content

Pull requests: antirez/ds4

New pull request New
169 Open 142 Closed
169 Open 142 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

GLM 5.2: CUDA backend support
#517 opened Jul 7, 2026 by giannisanni Loading…
Server: add --parallel N interleaved sessions (default 1)
#516 opened Jul 7, 2026 by iCreil Loading…
Engine: share prefill/decode work buffers across sessions
#515 opened Jul 7, 2026 by iCreil Loading…
Add sidecar for speed up ssd streaming
#514 opened Jul 7, 2026 by GiorgioOppo Loading…
cuda: fix uninitialized IQ2 dequant LUTs in batch MoE expert-tile kernels (n_embd > 4096)
#513 opened Jul 7, 2026 by giannisanni Loading…
port: Windows native ROCm build for AMD Strix Halo (gfx1151)
#512 opened Jul 7, 2026 by paperlinguist • Draft
Fix CPU backend abort on DeepSeek V4 PRO: use DS4_N_OUT_GROUP for attention-output grouping
#511 opened Jul 6, 2026 by mrighini Loading…
1
CPU backend: SSD streaming for routed experts (safe CPU inference on macOS)
#508 opened Jul 6, 2026 by Tatlatat Loading…
CUDA streaming: per-size-class expert caches, keep cache warm across prefills
#504 opened Jul 5, 2026 by iCreil Loading…
1
Reconcile DSpark runtime (#480) + B2 rejection sampling/adaptive block sizing (#482)
#502 opened Jul 5, 2026 by stephenlthorn Loading…
3
Opt-in mlock of non-routed weights for SSD streaming (DS4_MLOCK_NONROUTED)
#499 opened Jul 4, 2026 by hi-pauls Loading…
6 tasks done
2
rocm: use numerically stable SiLU to avoid NaN on large-magnitude gate
#498 opened Jul 4, 2026 by Flowers-of-Romance Loading…
2
cuda: run streaming selected load for single-token ffn batch encodes
#497 opened Jul 4, 2026 by giannisanni Loading…
cuda: support CUDA toolkits older than 12.2
#494 opened Jul 4, 2026 by giannisanni Loading…
Draft of Inference Sandbox jj-dai.org
#490 opened Jul 3, 2026 by VLADLEVIT • Draft
2
Server: fix agent-loop cache misses, add cancellation, observability, and robustness fixes
#489 opened Jul 2, 2026 by elkaix Loading…
cuda: enable streaming auto cache (implement recommended_working_set_size)
#488 opened Jul 2, 2026 by riccardo-galbani Loading…
1
cuda: fall back to pinned host memory when the model arena runs out of VRAM
#487 opened Jul 2, 2026 by riccardo-galbani Loading…
Enable MI300X ROCm support
#484 opened Jul 1, 2026 by ehartford Loading…
6
DSpark B2 rejection sampling + adaptive block sizing
#482 opened Jun 30, 2026 by machiabeli Loading…
6
feat: add headless browser support with curl fallback for web tools
#479 opened Jun 29, 2026 by J3rr1ck Loading…
CUDA: make DeepSeek-V4-Pro correct on the indexed-attention path (top_k 512→1024) + enable decode LUT gate for in_dim>4096
#478 opened Jun 29, 2026 by slackarea Loading…
Support DeepSeek V4 Flash 4Expert (top-4)
#474 opened Jun 28, 2026 by yuhai-china Loading…
CUDA: scale q8->f16 cache reserve on >=112 GiB cards (fixes session OOM on large models)
#472 opened Jun 28, 2026 by slackarea Loading…
Fix CUDA MoE router hardcoded to 256 experts
#466 opened Jun 27, 2026 by slackarea Loading…
ProTip! Filter pull requests by the default branch with base:main.