wangchongyi
commited on
Commit
·
d272cfd
1
Parent(s):
73cb2e7
add multi-image infer usage
Browse files
README.md
CHANGED
@@ -260,6 +260,35 @@ answer = model.chat(
|
|
260 |
print(answer)
|
261 |
```
|
262 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
263 |
|
264 |
## License
|
265 |
#### Model License
|
|
|
260 |
print(answer)
|
261 |
```
|
262 |
|
263 |
+
#### Chat with multiple images
|
264 |
+
<details>
|
265 |
+
<summary> Click to show Python code running MiniCPM-V 4.5 with multiple images input. </summary>
|
266 |
+
|
267 |
+
```python
|
268 |
+
import torch
|
269 |
+
from PIL import Image
|
270 |
+
from transformers import AutoModel, AutoTokenizer
|
271 |
+
|
272 |
+
model = AutoModel.from_pretrained('openbmb/MiniCPM-V-4_5', trust_remote_code=True,
|
273 |
+
attn_implementation='sdpa', torch_dtype=torch.bfloat16) # sdpa or flash_attention_2
|
274 |
+
model = model.eval().cuda()
|
275 |
+
tokenizer = AutoTokenizer.from_pretrained('openbmb/MiniCPM-V-4_5', trust_remote_code=True)
|
276 |
+
|
277 |
+
image1 = Image.open('image1.jpg').convert('RGB')
|
278 |
+
image2 = Image.open('image2.jpg').convert('RGB')
|
279 |
+
question = 'Compare image 1 and image 2, tell me about the differences between image 1 and image 2.'
|
280 |
+
|
281 |
+
msgs = [{'role': 'user', 'content': [image1, image2, question]}]
|
282 |
+
|
283 |
+
answer = model.chat(
|
284 |
+
image=None,
|
285 |
+
msgs=msgs,
|
286 |
+
tokenizer=tokenizer
|
287 |
+
)
|
288 |
+
print(answer)
|
289 |
+
```
|
290 |
+
</details>
|
291 |
+
|
292 |
|
293 |
## License
|
294 |
#### Model License
|