Official implementation of "Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding"
Python · active 2025-05-28 → 2026-05-30 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 368 days from 2025-05-28 to 2026-05-30. Pushes: 45 total, peak 4 in a day. Pull requests: 10 total, peak 1 in a day. Issues: 34 total, peak 3 in a day. Comments: 40 total, peak 3 in a day. Stars: 417 total, peak 25 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 |
|---|---|---|---|---|
| hills-code | 67 | 39 | 2 | 18 |
| xiwen1 | 6 | 6 | 0 | 0 |
| ptx9363 | 4 | 0 | 1 | 2 |
| preordinary | 3 | 0 | 0 | 2 |
| AndyJi1 | 3 | 0 | 0 | 1 |
| friedrichor | 3 | 0 | 0 | 1 |
| ParadoxZW | 3 | 0 | 0 | 2 |
| littlestone111 | 2 | 0 | 0 | 1 |
| Agrim-Jain | 2 | 0 | 0 | 1 |
| Lyn-Lucy | 2 | 0 | 0 | 1 |
| zhengli97 | 2 | 0 | 0 | 1 |
| lihe07 | 2 | 0 | 1 | 1 |
| yeonjoon-jung01 | 1 | 0 | 0 | 0 |
| Crys-Chen | 1 | 0 | 0 | 0 |
| yongtang2025 | 1 | 0 | 0 | 1 |
| Laurence-Wu | 1 | 0 | 0 | 0 |
| xiaoshideta | 1 | 0 | 0 | 1 |
| zachary19889 | 1 | 0 | 0 | 0 |
| NickCheng0921 | 1 | 0 | 1 | 0 |
| jiangzizi | 1 | 0 | 0 | 1 |
Recent activity
Latest issues, pull requests and releases
- Pull request#77hills-code2026-05-27 03:27
- Issue comment#62Tomingz2026-04-24 09:31[v2] Reproducibility: release paper training script, preprocessing, and exact hyperparameters
- Pull request#74SensenGao2026-04-17 01:38
- Issue#64Laurence-Wu2026-01-16 00:49batch size > 1 and temperature != 0.0 not supported by the generate function file
- Issue#55zhengli972026-01-05 07:19Error when changing the block size to 32
- Issue#62yeonjoon-jung012026-01-05 06:37[v2] Reproducibility: release paper training script, preprocessing, and exact hyperparameters
- Issue#61lizhuo0082025-12-30 15:56Why does Dream’s cache “global update” always update from current_block_start instead of using a sampling strategy?
- Issue comment#59yongtang20252025-12-30 08:49请问为什么在eval_gsm8k.sh中，dual cache+paralle的steps参数是传入length，而不是steps
- Issue comment#54littlestone1112025-12-29 16:22Question Regarding the Llada Parallel decoding in GSM8K
- Issue#59zqc2142025-12-29 05:02请问为什么在eval_gsm8k.sh中，dual cache+paralle的steps参数是传入length，而不是steps
- Issue comment#55zhengli972025-12-26 07:08Error when changing the block size to 32
- Issue comment#55hills-code2025-12-25 13:59Error when changing the block size to 32
- Issue comment#54hills-code2025-12-25 12:51Question Regarding the Llada Parallel decoding in GSM8K
- Issue comment#23AndyJi12025-12-25 09:07Does the evaluation of MBPP require postprocessing, as in HumanEval?
- Issue comment#53Agrim-Jain2025-12-19 02:47Unable to reproduce results for HumanEval dataset
- Issue#53Agrim-Jain2025-12-18 14:55Unable to reproduce results for HumanEval dataset
- Issue#52quaternior2025-12-11 05:24About warm-up stage and cache reuse stage in dual cache+parallel
- Pull request#51NickCheng09212025-12-10 06:51
- Issue#37hills-code2025-11-28 09:08Error when using the README code
- Issue#42hills-code2025-11-28 09:07parralel_question
- Issue comment#34hills-code2025-11-28 09:05fast-dllm-v2 mbpp and humaneval evaluation
- Issue#38hills-code2025-11-28 09:04Bug: Changing threshold in Fast-dLLM v2 MMLU Inference Has No Effect
- Issue comment#38hills-code2025-11-28 09:04Bug: Changing threshold in Fast-dLLM v2 MMLU Inference Has No Effect
- Issue comment#36lihe072025-11-27 09:14fast-dllm-v2中，dual cache的实际作用？
- Issue comment#20hills-code2025-11-25 13:11Dream parallel decoding implementation
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 417 stars here means stars gained during the window, not the repo's star count.