Robotics
LeRobot
Safetensors
reward-model
distributional_value_function
lilkm's picture
lilkm HF Staff
Upload reward model weights, train config and readme
a152b63 verified
Raw
History Blame Contribute Delete
1.83 kB
{
"type": "distributional_value_function",
"input_features": {
"observation.state": {
"type": "STATE",
"shape": [
6
]
},
"observation.images.top": {
"type": "VISUAL",
"shape": [
3,
480,
640
]
},
"observation.images.wrist": {
"type": "VISUAL",
"shape": [
3,
480,
640
]
}
},
"output_features": {},
"device": "cuda",
"pretrained_path": null,
"pretrained_revision": null,
"push_to_hub": true,
"repo_id": "lilkm/vf_stackblocks_recap_gemma3_v1",
"license": null,
"tags": null,
"private": null,
"siglip_path": "google/siglip2-so400m-patch14-384",
"gemma3_path": "google/gemma-3-270m",
"vlm_pretrained_path": null,
"num_value_bins": 201,
"value_support_min": -1.0,
"value_support_max": 0.0,
"hl_gauss_sigma_ratio": 0.75,
"target_method": "hl_gauss",
"use_one_hot_terminal": false,
"image_resolution": [
448,
448
],
"num_image_tokens": 256,
"tokenizer_max_length": 200,
"value_dropout": 0.1,
"freeze_vision_encoder": true,
"freeze_language_model": false,
"stop_gradient_to_vlm": false,
"optimizer_vision_lr": 1e-06,
"optimizer_language_model_lr": 1e-05,
"optimizer_multimodal_projector_lr": 5e-05,
"optimizer_value_query_lr": 0.0001,
"optimizer_value_head_lr": 0.0001,
"optimizer_weight_decay": 0.01,
"scheduler_warmup_steps": 500,
"scheduler_decay_steps": 12000,
"scheduler_decay_lr": 1e-06,
"vision_encoder_lr_multiplier": 0.5,
"normalization_mapping": {
"VISUAL": "IDENTITY"
}
}