Awomo-0.5
Collection
1 item • Updated
Awomo-0.5 for RoboDojo
This repository provides inference weights for the Awomo-0.5 RoboDojo release,
using the Awomo05 adapter in XPolicyLab. Access and use are subject to
the evaluation-only terms in LICENSE.md. This is not a grant of
permission to redistribute, train, fine-tune, or distill the model.
Download model.pt, dataset_stats.json, and SHA256SUMS together. Keep the
normalization statistics paired with this checkpoint and verify the files with
sha256sum -c SHA256SUMS.
Evaluation code is maintained in the official
XPolicyLab repository,
with plugin name Awomo05.
The RoboDojo upstream baseline is 2184bf8844ea9d205382c4aefa3a694311418251;
local evaluation used adjusted XPolicyLab submodule configuration.
Use input_color_order: bgr, action_type: joint, and replan_steps: 8.
The checkpoint requires separately obtained FLUX.2-klein-base-4B, FLUX.2-dev autoencoder, and Qwen3-VL-4B-Instruct base weights. Their licenses remain applicable. Follow the adapter README for dependency installation, base-weight paths, and policy-server startup.
Training data, task-specific annotations, and training metadata are not included.
- Downloads last month
- 34