Optimized DFlash: batch parallel, T2T, multiple steps denoise
Python · active 2026-01-06 → 2026-08-21 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 228 days from 2026-01-06 to 2026-08-21. Pushes: 30 total, peak 3 in a day. Pull requests: 4 total, peak 3 in a day. Issues: 30 total, peak 6 in a day. Comments: 51 total, peak 4 in a day. Stars: 736 total, peak 39 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 |
|---|---|---|---|---|
| jianc99 | 66 | 29 | 0 | 28 |
| shaun0927 | 4 | 0 | 3 | 0 |
| Sylvan820 | 3 | 0 | 0 | 2 |
| hongyanz | 2 | 0 | 0 | 1 |
| shx2005 | 2 | 0 | 0 | 0 |
| PlyMxt | 2 | 0 | 0 | 1 |
| jiahe7ay | 2 | 0 | 0 | 1 |
| Arcmoon-Hu | 2 | 0 | 0 | 2 |
| Ximingwang-09 | 2 | 0 | 0 | 2 |
| Ofir408 | 2 | 0 | 0 | 1 |
| kirilgorbachov | 1 | 0 | 0 | 1 |
| johanna-rock-tt | 1 | 0 | 0 | 0 |
| gaoxiangxiang1996 | 1 | 0 | 0 | 0 |
| arunpatala | 1 | 0 | 0 | 0 |
| alipourranjbar | 1 | 0 | 0 | 1 |
| maodoudou168 | 1 | 0 | 0 | 1 |
| Walid-Ahmed | 1 | 0 | 0 | 1 |
| mingMelody | 1 | 0 | 0 | 0 |
| pocca2048 | 1 | 0 | 0 | 0 |
| shreyas269 | 1 | 0 | 0 | 1 |
Recent activity
Latest issues, pull requests and releases
- Issue#118luolun2026-06-11 01:38Qwen3.5-35B-A3B Mean acceptance length low
- Issue comment#66Sylvan8202026-05-25 09:42Decode performance degrades significantly with long context
- Issue#123Sylvan8202026-05-19 11:43CUDA 12.8 install of dflash[sglang] fails
- Issue#121dongyibo2026-05-15 07:30DFLASH for GLM51
- Issue comment#101fritol2026-05-10 12:31Gemma4 DFlash Docker image on GHCR returns unauthorized
- Issue#98johanna-rock-tt2026-05-04 08:53Request: DFlash draft checkpoint for deepseek-ai/DeepSeek-R1-0528
- Issue comment#88PlyMxt2026-04-25 08:04Request: DFlash draft checkpoint for deepseek-ai/DeepSeek-V4-Flash
- Issue#88PlyMxt2026-04-24 06:04Request: DFlash draft checkpoint for deepseek-ai/DeepSeek-V4-Flash
- Issue#86xinanjiao2026-04-23 07:09Does the DFlash acceleration effect on Qwen3.5 2B or 0.8B show significant improvement?
- Pull request#84Jah-yee2026-04-21 19:28
- Pull request#76shaun09272026-04-17 08:45
- Issue#75jianc992026-04-17 08:43MLX: _GDNStateCapture.rollback assumes "non-trimmable cache count == GDN layer count" without an assertion
- Pull request#78shaun09272026-04-17 08:43
- Issue#74jianc992026-04-17 08:36dflash_generate: draft sampler ignores temperature; speculative decoding distribution diverges from target for temperature > 0
- Pull request#78shaun09272026-04-17 08:05
- Issue#74shaun09272026-04-17 07:37dflash_generate: draft sampler ignores temperature; speculative decoding distribution diverges from target for temperature > 0
- Issue comment#68jianc992026-04-17 07:03Request: DFlash draft checkpoint for Qwen3.6-35B-A3B
- Issue comment#66NASLXTO2026-04-16 23:19Decode performance degrades significantly with long context
- Issue comment#47meanaverage2026-04-16 05:19Dflash support for gemma4
- Issue comment#40kirilgorbachov2026-04-11 13:09Qwen3.5 27b and 122b
- Issue comment#47itzpingcat2026-04-08 01:52Dflash support for gemma4
- Issue comment#44jianc992026-04-04 10:02HF backend support for Qwen3.5 model family
- Issue#6jianc992026-03-26 00:36Support for GPT-OSS-120B and VLLM.
- Issue#20jianc992026-03-26 00:33Very slow on my nvidia 3090: 44 tok/s
- Issue#36jianc992026-03-26 00:30Crashing sometimes
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 736 stars here means stars gained during the window, not the repo's star count.