Update custom_st.py (#19)
Browse files- Update custom_st.py (89146e49f8e1cd1cd5231642a803167c3868f443)
- Update custom_st.py (23e2bf96c6f5d8b13f352d4ac29cd153f522ab91)
- Update custom_st.py (7a9a21cc961c2fd89de1c56c616a7c64e9cbcd2a)
Co-authored-by: kosung <[email protected]>
- custom_st.py +3 -1
custom_st.py
CHANGED
|
@@ -1,3 +1,5 @@
|
|
|
|
|
|
|
|
| 1 |
from io import BytesIO
|
| 2 |
from typing import Any, Dict, Optional, List
|
| 3 |
import torch
|
|
@@ -51,7 +53,7 @@ class MultiModalTransformer(BaseTransformer):
|
|
| 51 |
self, features: Dict[str, torch.Tensor], **kwargs
|
| 52 |
) -> Dict[str, torch.Tensor]:
|
| 53 |
if features.get("inputs_embeds", None) is None:
|
| 54 |
-
features["inputs_embeds"] = self.auto_model.base_model.
|
| 55 |
if features.get("pixel_values", None) is not None:
|
| 56 |
features["pixel_values"] = features["pixel_values"].type(self.auto_model.visual.get_dtype())
|
| 57 |
image_embeds = self.auto_model.visual(
|
|
|
|
| 1 |
+
import math
|
| 2 |
+
import logging
|
| 3 |
from io import BytesIO
|
| 4 |
from typing import Any, Dict, Optional, List
|
| 5 |
import torch
|
|
|
|
| 53 |
self, features: Dict[str, torch.Tensor], **kwargs
|
| 54 |
) -> Dict[str, torch.Tensor]:
|
| 55 |
if features.get("inputs_embeds", None) is None:
|
| 56 |
+
features["inputs_embeds"] = self.auto_model.base_model.get_input_embeddings()(features["input_ids"])
|
| 57 |
if features.get("pixel_values", None) is not None:
|
| 58 |
features["pixel_values"] = features["pixel_values"].type(self.auto_model.visual.get_dtype())
|
| 59 |
image_embeds = self.auto_model.visual(
|