[NeurIPS'24 Spotlight, ICLR'25] To speed up Long-context LLMs' inference, approximate and dynamic sparse calculate the attention, which reduces inference latency by up to 10x for pre-filling on an A100 while maintaining accuracy.
active 2025-03-14 → 2026-04-27 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 410 days from 2025-03-14 to 2026-04-27. Pushes: 18 total, peak 5 in a day. Pull requests: 31 total, peak 8 in a day. Issues: 24 total, peak 2 in a day. Comments: 62 total, peak 22 in a day. Stars: 120 total, peak 4 in a day.
- Pushes
- Pull requests
- Issues
- Comments
- Stars
Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.
Top contributors
Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity
| Contributor | Contributions | Pushes | PRs | Comments |
|---|---|---|---|---|
| iofu728 | 53 | 14 | 23 | 9 |
| Copilot | 50 | 0 | 0 | 31 |
| HalberdOfPineapple | 8 | 4 | 3 | 1 |
| microsoft-github-policy-service[bot] | 5 | 0 | 0 | 5 |
| zerolllin | 5 | 0 | 2 | 2 |
| GuoYiFantastic | 4 | 0 | 1 | 0 |
| xwpaul3 | 4 | 0 | 0 | 1 |
| Andy0422 | 3 | 0 | 0 | 1 |
| copilot-pull-request-reviewer[bot] | 2 | 0 | 0 | 0 |
| gaowayne | 2 | 0 | 0 | 1 |
| Enoch202 | 2 | 0 | 0 | 0 |
| fluctlux | 2 | 0 | 0 | 1 |
| lepangdan | 2 | 0 | 0 | 0 |
| Ewan-Ding | 1 | 0 | 0 | 0 |
| wagnzi | 1 | 0 | 0 | 0 |
| ponytaill | 1 | 0 | 0 | 1 |
| KevinWu2017 | 1 | 0 | 0 | 0 |
| yanghailong-git | 1 | 0 | 0 | 1 |
| RuixiangMa | 1 | 0 | 1 | 0 |
| testpppppp | 1 | 0 | 0 | 0 |
Recent activity
Latest issues, pull requests and releases
- Pull request#206dependabot[bot]2026-03-09 02:59
- Pull request#205HalberdOfPineapple2026-03-09 02:58
- Issue comment#205microsoft-github-policy-service[bot]2026-02-08 14:45Feature(MTraining): add MTraining
- Pull request#205HalberdOfPineapple2026-02-08 14:44
- Issue comment#204HalberdOfPineapple2026-02-08 14:43Merge MTraining into MInference
- Pull request#204HalberdOfPineapple2026-02-08 14:41
- Issue#202wagnzi2026-01-01 07:51[Question]: Scbench, mulit turn, input=last turn true answers + now turn prompt, but cache is only the 1st turn prompt?????
- Issue#199KevinWu20172025-12-02 11:35[Question]: Out of memory for MInference in e2e benchmark test & NIAH with long context_window
- Pull request#198RuixiangMa2025-11-28 09:36
- Issue#197paitsai2025-11-25 18:47[Question]: Broken link in `Leank`: booksum.jsonl.zst dataset URL is inaccessible
- Issue#192xwpaul32025-10-20 08:30[Question]: Does benchmark_e2e_vllm_tp support TriMix now?
- Issue comment#186zZzZ9zZ92025-10-05 10:04[Feature Request]: Can MInference support chunked prefill?
- Issue#187iofu7282025-09-30 12:43[Bug]: cannot import name 'patch_leank' from 'minference.patch'
- Pull request#189iofu7282025-09-30 12:43
- Pull request#189zerolllin2025-09-30 10:26
- Issue comment#186KevinZeng082025-09-28 07:03[Feature Request]: Can MInference support chunked prefill?
- Issue#188Andy04222025-09-25 09:03[Question]: No speedup on multi-card with vllm
- Issue#187MamieZhu2025-09-25 02:04[Bug]: cannot import name 'patch_leank' from 'minference.patch'
- Issue#186zongy172025-09-19 02:53[Feature Request]: Can MInference support chunked prefill?
- Issue comment#178yanghailong-git2025-09-16 08:36[Bug]: Failed to launch vllm while import MInference
- Issue comment#176xwpaul32025-09-15 09:06[Question]: Implement triangle attention in sglang?
- Issue#185xwpaul32025-09-15 08:50[Question]: how to use inference in sglang
- Issue comment#183hzy462025-09-08 07:42[Bug]: Implement of TriangleMix
- Issue comment#175wangshang232025-09-05 09:07[Question]: Is MMinference included?
- Issue#182Enoch2022025-09-01 08:27[Bug]: Unable to run streamingLLM properly on minference with the qwen3-0.6B model.
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 120 stars here means stars gained during the window, not the repo's star count.