Code to pretrain, fine-tune, and evaluate DreamZero and run sim & real-world evals
Python · active 2026-01-28 → 2026-08-07 (UTC)
Activity over time
Daily event counts in the loaded window
Line chart, 192 days from 2026-01-28 to 2026-08-07. Pushes: 27 total, peak 7 in a day. Pull requests: 6 total, peak 2 in a day. Issues: 18 total, peak 2 in a day. Comments: 15 total, peak 2 in a day. Stars: 494 total, peak 98 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 |
|---|---|---|---|---|
| joeljang | 19 | 12 | 1 | 4 |
| seonghyeonye | 10 | 6 | 0 | 2 |
| NValec | 7 | 4 | 2 | 1 |
| zchuning | 5 | 5 | 0 | 0 |
| persuer0 | 2 | 0 | 0 | 0 |
| Undefinedx-Zhang | 2 | 0 | 0 | 1 |
| Dingrui-Wang | 2 | 0 | 0 | 1 |
| haosenwang1018 | 1 | 0 | 1 | 0 |
| Chunchunyoyo | 1 | 0 | 0 | 1 |
| magic097 | 1 | 0 | 0 | 1 |
| suixin1424 | 1 | 0 | 0 | 1 |
| 2460707925 | 1 | 0 | 0 | 0 |
| ShuaiZhou302 | 1 | 0 | 0 | 0 |
| Loki-Lu | 1 | 0 | 0 | 0 |
| boqian-li | 1 | 0 | 0 | 0 |
| whyxz | 1 | 0 | 0 | 0 |
| jha-manav | 1 | 0 | 0 | 0 |
| parrot18 | 1 | 0 | 1 | 0 |
| CharioteerStark | 1 | 0 | 1 | 0 |
| cwchenwang | 1 | 0 | 0 | 1 |
Recent activity
Latest issues, pull requests and releases
- Pull request#77CharioteerStark2026-06-04 22:32
- Issue#7224607079252026-05-24 09:52Question on 1-Step Inference Logic in DreamZero-Flash
- Issue comment#56suixin14242026-05-22 12:44How long will it take for finetune?
- Pull request#69parrot182026-05-01 12:32
- Issue#63DafengChi2026-04-20 03:02Wan2.1-I2V-5B-480P
- Issue#46seonghyeonye2026-04-19 06:52Question: Why does `test_client_AR.py` send 4 frames per chunk instead of 8?
- Issue comment#46seonghyeonye2026-04-19 06:52Question: Why does `test_client_AR.py` send 4 frames per chunk instead of 8?
- Issue comment#42seonghyeonye2026-04-19 06:41[Question] Possible incorrect architecture specs for Wan2.1-I2V-14B in comparison table
- Issue#32seonghyeonye2026-04-19 06:26Clarification on 7Hz inference frequency: per-chunk or per-action?
- Issue#61boqian-li2026-04-16 18:29State token attention mismatch between training and inference
- Issue#58ShuaiZhou3022026-04-15 06:00Training steps / schedule for new-embodiment (YAM) fine-tuning in the paper vs. released scripts
- Issue#50Undefinedx-Zhang2026-04-01 03:03Question regarding the specific benchmarks used on the MolmoSpaces leaderboard
- Issue comment#31Undefinedx-Zhang2026-04-01 02:58Any plan to release a flash-version of 5B checkpoint?
- Issue comment#449293622642026-03-25 01:34About how to identify the instruction is done
- Issue comment#43NVshawnhwang2026-03-24 22:48error with batch size > 1
- Issue comment#34NValec2026-03-18 22:18no experiment_cfg in AgiBot ckpt
- Issue#21joeljang2026-03-18 18:24How to generate longer predictive video?
- Issue#35joeljang2026-03-18 18:19How to set up simulation-based evaluation baselines without real robot hardware?
- Issue#34Megan-Jiang2026-03-17 06:40no experiment_cfg in AgiBot ckpt
- Issue#31whyxz2026-03-16 06:35Any plan to release a flash-version of 5B checkpoint?
- Issue comment#26joeljang2026-03-10 23:21Fix Sim Eval Error
- Issue#25Loki-Lu2026-03-10 02:54RuntimeError: Broadcasting dimension mismatch in action loss computation when batch_size > 1
- Issue comment#19Chunchunyoyo2026-03-05 06:23flash-attn can only support torch 2.5cxx max
- Issue comment#17joeljang2026-03-01 07:18DREAMZERO-AgiBot checkpoint
- Issue#19persuer02026-02-28 06:24flash-attn can only support torch 2.5cxx max
Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 494 stars here means stars gained during the window, not the repo's star count.