Instructions to use zhuyuze/zero-wam-ckpts with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use zhuyuze/zero-wam-ckpts with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("zhuyuze/zero-wam-ckpts", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("zhuyuze/zero-wam-ckpts", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]zhuyuze/zero-wam-ckpts
Zero-WAM checkpoints moved between servers with zw_hf.py (Zero-WAM git: script/hf_ckpt/). A component whose tensors, header metadata and configs equal those of a public base or of another entry is stored once and comes back as a symlink; the loaded weights are bit-identical.
hf download zhuyuze/zero-wam-ckpts --include '_tools/*' --local-dir zw_hf # or git pull: script/hf_ckpt/zw_hf.py
python zw_hf/_tools/zw_hf.py env # token, network, disk
python zw_hf/_tools/zw_hf.py pull <entry|alias> # + bases, links, sha256 check
Runbook for the agent on the eval server: _tools/REMOTE_AGENT.md.
| entry | aliases | kind | stored here | full size | base | step | server module | note |
|---|---|---|---|---|---|---|---|---|
1028_0117_HFS_own24_m48_pt_step2000 |
HFS_pt_s2000 | delta | 0.45 GB | 37.52 GB | zero-wam-posttrain-robotwin | 2000 | research.headft.server | own24 (video) + m48 (action) head deltas, split loss, from posttrain, step 2000; A+B 90/139 (64.7 %), set C 71/139; posttrain stock A+B 42/139 |
1029_1154_HFS_m48_pt_step2000 |
HFS_m48_pt_s2000 | delta | 0.30 GB | 37.37 GB | zero-wam-posttrain-robotwin | 2000 | research.headft.server | m48 action-head deltas alone (split recipe), from posttrain, step 2000; A+B 78/139 |
1030_0052_FT_full_pt_step2000 |
FT_full_pt_s2000 | full | 22.86 GB | 37.06 GB | zero-wam-posttrain-robotwin | 2000 | full FT from posttrain, A+B 70/137 (51.1 %) | |
1030_0054_HF_own24_pt_step2000 |
HF_own24_pt_s2000 | delta | 0.15 GB | 37.22 GB | zero-wam-posttrain-robotwin | 2000 | research.headft.server | own24 video-head deltas, plain-sum loss, from posttrain, step 2000; A+B 71/137 |
Public bases (downloaded from their own repos):
zero-wam-posttrain-robotwin:robbyant-research/zero-wam-posttrain-robotwin@07ee865f175d9474a5653e9147698390e47b6143
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support