Skip to content

Pull requests: microsoft/onnxruntime-genai

New pull request New
61 Open 1,671 Closed
61 Open 1,671 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

Add prefix-aware paged KV caching to the continuous-batching engine
#2420 opened Aug 15, 2026 by Gopalakrishnan Nallasamy (GopalakrishnanN) • Draft
4
Add recompute preemption to the continuous-batching engine
#2418 opened Aug 15, 2026 by Gopalakrishnan Nallasamy (GopalakrishnanN) • Draft
Retry smaller plans after execution capacity failures
#2417 opened Aug 14, 2026 by Baiju Meswani (baijumeswani) Collaborator Loading…
3
Add documentation for speculative decoding (base + n-gram)
#2414 opened Aug 14, 2026 by Samyak (samsat701) Contributor • Draft
Add speculative decoding (n-gram + base) benchmarking scripts
#2413 opened Aug 13, 2026 by Samyak (samsat701) Contributor • Draft
Add E2E Tool Calling Support to GenAI Engine
#2412 opened Aug 13, 2026 by Sayan Shaw (sayanshaw24) Collaborator • Draft
Add a native Gemma 3n multimodal processor
#2410 opened Aug 13, 2026 by NishantN2005 Loading…
5
Fix Android 16 KB page alignment for libmat.so (1DS telemetry)
#2406 opened Aug 12, 2026 by sheetalarkadam Contributor • Draft
Validate static batching configuration
#2403 opened Aug 12, 2026 by danielsongmicrosoft Contributor Loading…
6
Fix malformed multi-image grid test input
#2400 opened Aug 12, 2026 by Akshay Sonawane (apsonawane) Contributor Loading…
1
Add Muse Glimmer multimodal runtime type
#2397 opened Aug 11, 2026 by Justin Chu (justinchuby) Contributor Loading…
7
Optimize paged-attention logits memory usage
#2393 opened Aug 10, 2026 by Gopalakrishnan Nallasamy (GopalakrishnanN) Loading…
12
Increase MTP prefill chunk default to 1024 model:qwen-3.6
#2391 opened Aug 10, 2026 by Tianlei Wu (tianleiwu) Contributor Loading…
2
Qwen3.6: fuse shared-expert gating with GatedAdd model:qwen-3.6
#2390 opened Aug 10, 2026 by Tianlei Wu (tianleiwu) Contributor Loading…
Qwen3.6 MTP: share the embedding and NVFP4 LM head with the main decoder model:qwen-3.6
#2389 opened Aug 9, 2026 by Tianlei Wu (tianleiwu) Contributor Loading…
Use MRotaryEmbedding contrib op for Qwen2.5-VL/Qwen3-VL M-RoPE instead of manual graph construction
#2386 opened Aug 8, 2026 by Copilot AI • Draft
@kunal-vaishnavi
2
Bump the github-actions group across 1 directory with 14 updates dependencies Pull requests that update a dependency file github_actions Pull requests that update GitHub Actions code
#2385 opened Aug 6, 2026 by dependabot Bot Loading…
2
Extend search.chunk_size prefill chunking to the multimodal (VLM) pipeline
#2378 opened Aug 5, 2026 by Copilot AI • Draft
1
@kunal-vaishnavi
1
Add multimodal (VLM/ALM) benchmarking to the C++ model_benchmark
#2377 opened Aug 5, 2026 by Copilot AI • Draft
1
@kunal-vaishnavi
1
feat: AMD GPU umbrella execution provider integration
#2376 opened Aug 5, 2026 by Wang, Zhenze (zz002) • Draft
2 tasks done
1
Fix Qwen3.5 shared-expert export node names
#2375 opened Aug 5, 2026 by Yen-Shi Wang (yen-shi) Contributor Loading…
5
webgpu: Shape-free synthetic position_ids for Qwen3.5 VL graph capture
#2370 opened Aug 4, 2026 by Jiajia Qin (qjia7) Contributor • Draft
4 tasks
Integrating DeepSeek V4 Flash into ORT GenAI
#2366 opened Aug 3, 2026 by Copilot AI • Draft
@kunal-vaishnavi
5
Serve sliding-window paged layers from a ring of blocks (builder)
#2358 opened Jul 31, 2026 by Tianlei Wu (tianleiwu) Contributor • Draft
5
Serve sliding-window paged layers from a ring of blocks (runtime)
#2357 opened Jul 31, 2026 by Tianlei Wu (tianleiwu) Contributor Loading…
10
ProTip! Find all pull requests that aren't related to any open issues with -linked:issue.