DM05-libero is the LIBERO fine-tuned checkpoint of DM0.5, Dexmal's open-world Vision-Language-Action foundation model for embodied intelligence. DM0.5 uses a Gemma3 4B vision-language backbone with a 680M Action Expert to generate continuous robot actions, and is designed for natural-language manipulation, zero-shot generalization, efficient downstream fine-tuning, long-horizon historical context, robust policy behavior, and transfer across robot embodiments.

We recommend using Docker to set up the runtime environment first, which helps avoid version mismatches across CUDA, PyTorch, flash-attn, and other dependencies on the host machine.

Use the LIBERO-specific experiment configuration when evaluating this checkpoint. The inference service and benchmark client run separately and communicate over HTTP. When possible, use one GPU for the inference service and another for benchmark evaluation.

For the complete training and evaluation workflow, see the DM05 LIBERO Training and Evaluation Guide.

Run this command from the OpenDM repository root. The checkpoint must contain the matching norm_stats.json.

Keep the service running while executing the benchmark.

Edit evaluation/configs/libero/example_dm05_libero.yaml so that base_url points to the inference service. Select one of libero_spatial, libero_goal, libero_object, or libero_10 as the benchmark:

From the dexbotic-benchmark repository root, run:

Evaluation artifacts are written under the configured output_dir, including results.json, config.yaml, logs/evaluation.log, and rollout videos under videos/*.mp4.

We will continue to release more model weights, technical documentation, and examples. If this project is helpful to you, please consider giving us a star on GitHub

. Your support helps us move forward.