Bariona/fact-wam

FACT — RoboTwin Checkpoints

Action-transformer checkpoint for FACT (Failure-Aware Causal Training for World-Action Models), fine-tuned from Wan2.2-TI2V-5B on the RoboTwin 2.0 benchmark (data).

Paper: FACT: Failure-Aware Causal Training for World-Action Models
Project page: https://fact-wam.github.io
Code: https://github.com/Bariona/FACT

Checkpoints

Variant	Training data	Transformer directory
FACT (failure-cotrained)	Success and failure data	transformer/
Success-only	Success-only data	success_only/transformer/

Usage

The repo ships the matching normalization stats (norm_stats_delta.json), so inference/eval needs no dataset download or training:

huggingface-cli download Bariona/fact-wam --local-dir ./models/fact-wam

python -m scripts.inference_server \
  --model_id ./models/Wan2.2-TI2V-5B-Diffusers \
  --transformer_path ./models/fact-wam/transformer \
  --stats_path ./models/fact-wam/norm_stats_delta.json \
  --port 8093

For closed-loop RoboTwin evaluation, set TRANSFORMER_PATH=./models/fact-wam/transformer and STATS_PATH=./models/fact-wam/norm_stats_delta.json in evaluation/robotwin/launch_config.yml.

Success-only checkpoint

This checkpoint is trained on success-only data.

To download the success-only checkpoint and the normalization stats:

hf download Bariona/fact-wam \
  --include "success_only/*" "norm_stats_delta.json" \
  --local-dir ./models/fact-wam

Use --transformer_path ./models/fact-wam/success_only/transformer in the inference command above. For RoboTwin evaluation, set TRANSFORMER_PATH=./models/fact-wam/success_only/transformer. Use ./models/fact-wam/norm_stats_delta.json for --stats_path / STATS_PATH.