Robotics
Transformers
Safetensors
minicpm_vla
feature-extraction
vision-language-action
embodied-ai
minicpm
manipulation
custom_code
Instructions to use openbmb/MiniCPM-RobotManip with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use openbmb/MiniCPM-RobotManip with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("openbmb/MiniCPM-RobotManip", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
File size: 1,162 Bytes
7f5d27f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 | from __future__ import annotations
from transformers import PretrainedConfig
class MiniCPMVLAConfig(PretrainedConfig):
"""Configuration for the complete VLM and action-head VLA model."""
model_type = "minicpm_vla"
is_composition = True
def __init__(
self,
vlm_config: dict | PretrainedConfig | None = None,
action_dim: int = 80,
state_dim: int = 80,
action_horizon: int = 30,
max_num_embodiments: int = 32,
num_inference_timesteps: int = 4,
vlm_dtype: str = "bfloat16",
action_head_dtype: str = "float32",
**kwargs,
):
super().__init__(**kwargs)
if isinstance(vlm_config, PretrainedConfig):
vlm_config = vlm_config.to_dict()
self.vlm_config = vlm_config
self.action_dim = action_dim
self.state_dim = state_dim
self.action_horizon = action_horizon
self.max_num_embodiments = max_num_embodiments
self.num_inference_timesteps = num_inference_timesteps
self.vlm_dtype = vlm_dtype
self.action_head_dtype = action_head_dtype
self.architectures = ["MiniCPMV_VLA"]
|