from transformers import LlamaConfig | |
class LlamaActionConfig(LlamaConfig): | |
model_type = "llama_action" | |
def __init__(self, **kwargs): | |
super().__init__(**kwargs) | |
self.num_spatio_embeddings = kwargs.get("num_spatio_embeddings", 582) | |
self.num_temporal_embeddings = kwargs.get("num_temporal_embeddings", 25) | |
self.num_action_embeddings = kwargs.get("num_action_tokens", 5) | |
self.num_image_patches = kwargs.get("num_image_patches", 576) | |
self.action_dim = kwargs.get("action_dim", 3) | |