Update modeling_videollama3.py
Browse files- modeling_videollama3.py +1 -1
modeling_videollama3.py
CHANGED
|
@@ -286,7 +286,7 @@ class Videollama3MetaForCausalLM(ABC):
|
|
| 286 |
|
| 287 |
# 2. embed visual tokens
|
| 288 |
batched_num_patches = grid_sizes.prod(dim=1).div(merge_sizes ** 2).long()
|
| 289 |
-
mm_features = self.encode_images(pixel_values, grid_sizes, merge_sizes)
|
| 290 |
mm_features = self._get_valid_visual_tokens(mm_features, batched_num_patches, modals)
|
| 291 |
|
| 292 |
compression_mask = self._get_compression_mask(
|
|
|
|
| 286 |
|
| 287 |
# 2. embed visual tokens
|
| 288 |
batched_num_patches = grid_sizes.prod(dim=1).div(merge_sizes ** 2).long()
|
| 289 |
+
mm_features = self.encode_images(pixel_values, grid_sizes, merge_sizes).to(input_ids.device)
|
| 290 |
mm_features = self._get_valid_visual_tokens(mm_features, batched_num_patches, modals)
|
| 291 |
|
| 292 |
compression_mask = self._get_compression_mask(
|