Instructions to use jinaai/jina-bert-flash-implementation with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use jinaai/jina-bert-flash-implementation with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("jinaai/jina-bert-flash-implementation", trust_remote_code=True, dtype="auto") - Notebooks
- Google Colab
- Kaggle
feat: added cleaved_layers property
Browse files- modeling_bert.py +9 -0
modeling_bert.py
CHANGED
|
@@ -177,6 +177,15 @@ class BertEncoder(nn.Module):
|
|
| 177 |
assert 0 <= idx < len(self.layers)
|
| 178 |
self._last_layer_idx = idx
|
| 179 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 180 |
@property
|
| 181 |
def gradient_checkpointing(self):
|
| 182 |
return self._grad_checkpointing
|
|
|
|
| 177 |
assert 0 <= idx < len(self.layers)
|
| 178 |
self._last_layer_idx = idx
|
| 179 |
|
| 180 |
+
@property
|
| 181 |
+
def cleaved_layers(self):
|
| 182 |
+
return len(self.layers) - self.last_layer_idx - 1
|
| 183 |
+
|
| 184 |
+
@cleaved_layers.setter
|
| 185 |
+
def cleaved_layers(self, n: int):
|
| 186 |
+
assert 0 <= n < len(self.layers)
|
| 187 |
+
self.last_layer_idx = len(self.layers) - n - 1
|
| 188 |
+
|
| 189 |
@property
|
| 190 |
def gradient_checkpointing(self):
|
| 191 |
return self._grad_checkpointing
|