YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
MolmoAct2 VLA-Arena 训练超参数
1. Common
| Key | Value |
|---|---|
| action_mode | continuous |
| inference_action_mode | continuous |
| setup_type | single franka robotic arm |
| control_mode | delta end-effector pose |
| chunk_size | 10 |
| n_action_steps | 10 |
| n_obs_steps | 1 |
| num_flow_timesteps | 8 |
| image_keys | ["observation.images.image", "observation.images.wrist_image"] |
2. Dataset
| Key | Value |
|---|---|
| repo_id | vla-arena_v30 |
| video_backend | pyav |
| image_transforms.enable | true |
3. Model
| Key | Value |
|---|---|
| type | molmoact2 |
| model_dtype | bfloat16 |
| gradient_checkpointing | true |
| freeze_embedding | true |
| enable_lora_vlm | false |
| train_action_expert_only | false |
4. Training
| Key | Value |
|---|---|
| batch_size | 32 |
| steps | 10000 |
| optimizer_lr | 1e-5 |
| optimizer_vit_lr | 5e-6 |
| optimizer_connector_lr | 5e-6 |
| optimizer_action_expert_lr | 5e-5 |
| optimizer_grad_clip_norm | 1.0 |
| scheduler_warmup_steps | 500 |
| seed | 1000 |
5. System
| Key | Value |
|---|---|
| device | cuda |
| num_workers | 8 |
| eval_freq | -1 |
| log_freq | 10 |
| save_checkpoint | true |
| save_freq | 1500 |
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support