Commit
·
bd67535
1
Parent(s):
33f9417
switch back to opt-2.7b
Browse files
app.py
CHANGED
|
@@ -7,12 +7,7 @@ from PIL import Image
|
|
| 7 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 8 |
|
| 9 |
|
| 10 |
-
|
| 11 |
-
processor = AutoProcessor.from_pretrained("Salesforce/blip-itm-large-coco")
|
| 12 |
-
# Load model in int8 using bitsandbytes, and pass device_map='auto'
|
| 13 |
-
model = Blip2ForConditionalGeneration.from_pretrained(
|
| 14 |
-
"Salesforce/blip-itm-large-coco", load_in_8bit=True, device_map='auto'
|
| 15 |
-
)
|
| 16 |
|
| 17 |
|
| 18 |
# Load the BLIP-2 model and processor (needs A10G)
|
|
@@ -22,14 +17,14 @@ model = Blip2ForConditionalGeneration.from_pretrained(
|
|
| 22 |
# "Salesforce/blip2-opt-6.7b-coco", load_in_8bit=True, device_map='auto'
|
| 23 |
#)
|
| 24 |
|
| 25 |
-
# Uncomment lines 20, 22, 23, & 24 to begin using blip2-
|
| 26 |
|
| 27 |
# Load the BLIP-2 model and processor
|
| 28 |
-
|
| 29 |
# Load model in int8 using bitsandbytes, and pass device_map='auto'
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
|
| 34 |
def blip2_interface(image, prompted_caption_text, vqa_question, chat_context):
|
| 35 |
# Prepare image input
|
|
|
|
| 7 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 8 |
|
| 9 |
|
| 10 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
|
| 13 |
# Load the BLIP-2 model and processor (needs A10G)
|
|
|
|
| 17 |
# "Salesforce/blip2-opt-6.7b-coco", load_in_8bit=True, device_map='auto'
|
| 18 |
#)
|
| 19 |
|
| 20 |
+
# Uncomment lines 20, 22, 23, & 24 to begin using blip2-opt-2.7b model (can run on T4 Medium)
|
| 21 |
|
| 22 |
# Load the BLIP-2 model and processor
|
| 23 |
+
processor = AutoProcessor.from_pretrained("Salesforce/blip2-opt-2.7b")
|
| 24 |
# Load model in int8 using bitsandbytes, and pass device_map='auto'
|
| 25 |
+
model = Blip2ForConditionalGeneration.from_pretrained(
|
| 26 |
+
"Salesforce/blip2-opt-2.7b", load_in_8bit=True, device_map='auto'
|
| 27 |
+
)
|
| 28 |
|
| 29 |
def blip2_interface(image, prompted_caption_text, vqa_question, chat_context):
|
| 30 |
# Prepare image input
|