Robotics
Transformers
Safetensors
English
molmoact
image-text-to-text
molmo
olmo
reasoning
vla
manipulation
custom_code
Instructions to use allenai/MolmoAct-7B-D-0812 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use allenai/MolmoAct-7B-D-0812 with Transformers:
# Load model directly from transformers import AutoModelForImageTextToText model = AutoModelForImageTextToText.from_pretrained("allenai/MolmoAct-7B-D-0812", trust_remote_code=True, dtype="auto") - Notebooks
- Google Colab
- Kaggle
remove normalization of gripper state
Browse files- modeling_molmoact.py +2 -0
modeling_molmoact.py
CHANGED
|
@@ -1955,6 +1955,8 @@ class MolmoActForActionReasoning(MolmoActPreTrainedModel, GenerationMixin):
|
|
| 1955 |
q01 = np.asarray(stats["q01"], dtype=np.float32)
|
| 1956 |
q99 = np.asarray(stats["q99"], dtype=np.float32)
|
| 1957 |
mask = np.asarray(stats.get("mask", np.ones_like(q01, dtype=bool)), dtype=bool)
|
|
|
|
|
|
|
| 1958 |
|
| 1959 |
# Lazily load the tokenizer (shared across calls)
|
| 1960 |
if self._qwen_tokenizer is None:
|
|
|
|
| 1955 |
q01 = np.asarray(stats["q01"], dtype=np.float32)
|
| 1956 |
q99 = np.asarray(stats["q99"], dtype=np.float32)
|
| 1957 |
mask = np.asarray(stats.get("mask", np.ones_like(q01, dtype=bool)), dtype=bool)
|
| 1958 |
+
# the gripper state should not be normalized
|
| 1959 |
+
mask[-1] = False
|
| 1960 |
|
| 1961 |
# Lazily load the tokenizer (shared across calls)
|
| 1962 |
if self._qwen_tokenizer is None:
|