轻量级vLLM实现，从零开始构建的高效推理模型。 极速离线推理，性能与vLLM相当；代码简洁易读，仅1200行Python代码；多项优化技术，如前缀缓存、Torch编译等
Python · active 2025-06-12 → 2026-08-07 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 422 days from 2025-06-12 to 2026-08-07. Pushes: 25 total, peak 6 in a day. Pull requests: 56 total, peak 3 in a day. Issues: 70 total, peak 6 in a day. Comments: 99 total, peak 4 in a day. Stars: 5,919 total, peak 425 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 |
|---|---|---|---|---|
| GeeeekExplorer | 71 | 25 | 7 | 16 |
| alexhegit | 9 | 0 | 0 | 3 |
| Copilot | 7 | 0 | 0 | 2 |
| okigan | 5 | 0 | 2 | 3 |
| sudhir2016 | 4 | 0 | 0 | 1 |
| guoqingbao | 4 | 0 | 1 | 2 |
| xiaohajiayou | 3 | 0 | 2 | 1 |
| n1teshy-2 | 3 | 0 | 0 | 1 |
| skyloevil | 3 | 0 | 1 | 2 |
| tiannuo-yang | 3 | 0 | 1 | 1 |
| tpoisonooo | 3 | 0 | 3 | 0 |
| CzealChen | 3 | 0 | 1 | 2 |
| billishyahao | 3 | 0 | 1 | 1 |
| MasterGodzilla | 3 | 0 | 0 | 3 |
| DestineG | 3 | 0 | 0 | 1 |
| Terry-Uv | 3 | 0 | 1 | 2 |
| KKKZOZ | 2 | 0 | 0 | 2 |
| iamorangedad | 2 | 0 | 2 | 0 |
| Seas0 | 2 | 0 | 0 | 2 |
| jiqiujia | 2 | 0 | 0 | 2 |
Recent activity
Latest issues, pull requests and releases
- Issue comment#220NilsGraf2026-05-06 19:37[Feature Request] Support weightless RMSNorm (for FlashNorm weight folding trick)
- Issue#221Rising03212026-05-01 07:15[Discussion] As vLLM-Omni is also complex, I tried building a minimal version (~1k LOC) inspired by nano-vllm
- Issue comment#199GeeeekExplorer2026-04-27 06:33fix(engine): resolve crash on fully cached (100% match) prompt via L-1 fallback strategy
- Pull request#213Anai-Guo2026-04-25 09:43
- Issue comment#213GeeeekExplorer2026-04-25 09:43fix(model_runner): correct seqlen_k to chunk boundary in prepare_prefill
- Issue#208DestineG2026-04-23 04:28Prefix cache can prematurely hit in-flight blocks in same prefill batch
- Issue#212DestineG2026-04-23 04:27prepare_prefill logic flaw: seqlen_k incorrectly uses total sequence length during Chunked Prefill
- Issue comment#207DestineG2026-04-22 07:39fix(scheduler): recalculate num_tokens after allocate to prevent IndexError at model_runner:155
- Pull request#210tj12352026-04-22 04:26
- Issue#206Wangbiao22026-04-20 07:52Implement support for Qwen3.5
- Issue comment#144fishAndShrimp2026-04-13 18:34Two bugs
- Pull request#200KinglittleQ2026-04-13 13:13
- Issue comment#185GeeeekExplorer2026-04-13 13:08fix: update download command for model weights in README
- Pull request#172IceCreamMilkyTea2026-04-13 12:45
- Issue#154GeeeekExplorer2026-04-12 06:25Suggestion to add a standard attention implementation
- Issue comment#197jakezai250808-sudo2026-04-09 06:05docs: add source code study notes and learning resources
- Issue comment#189Barney6662026-04-02 04:13Qwen models initialization fails with TypeError: unhashable type: 'dict' and AssertionError due to rope_scaling
- Issue comment#153pangkun2482026-04-01 15:13fix: clean up hash_to_block_id mapping when deallocating blocks
- Issue comment#167Seas02026-03-24 04:39rope argument inconsistent
- Issue comment#189Seas02026-03-23 02:50Qwen models initialization fails with TypeError: unhashable type: 'dict' and AssertionError due to rope_scaling
- Pull request#186liulanze2026-03-18 04:43
- Pull request#180fly19892026-03-07 09:38
- Pull request#178pacoxu2026-03-03 03:12
- Issue comment#163jerryjedi2026-02-17 15:56Stale hash mapping remains in `hash_to_block_id` when reallocating blocks in `BlockManager` [Bug]
- Issue comment#167banfeb2026-02-15 09:27rope argument inconsistent
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 5,919 stars here means stars gained during the window, not the repo's star count.