Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import pipeline, BitsAndBytesConfig
|
2 |
+
import torch
|
3 |
+
|
4 |
+
quantization_config = BitsAndBytesConfig(
|
5 |
+
load_in_4bit=True,
|
6 |
+
bnb_4bit_compute_dtype=torch.bfloat16
|
7 |
+
)
|
8 |
+
|
9 |
+
# 13B version can be loaded with llava-hf/llava-1.5-13b-hf
|
10 |
+
model_id = "llava-hf/llava-1.5-7b-hf"
|
11 |
+
|
12 |
+
|
13 |
+
pipe = pipeline("image-to-text", model=model_id, model_kwargs={"quantization_config": quantization_config})
|
14 |
+
|
15 |
+
prompt = "USER: <image>\nDescribe this picture\nASSISTANT: "
|
16 |
+
|
17 |
+
outputs = pipe(image, prompt=prompt, generate_kwargs={"max_new_tokens": 200})
|
18 |
+
print(outputs[0]['generated_text'])
|
19 |
+
|
20 |
+
def update_conversation(new_message, history, image):
|
21 |
+
|
22 |
+
if image is None:
|
23 |
+
return "Please upload an image first using the widget on the left"
|
24 |
+
|
25 |
+
conversation_starting_from_image = [[user, assistant] for [user, assistant] in history if not assistant.startswith('Please')]
|
26 |
+
|
27 |
+
prompt = "USER: <image>\n"
|
28 |
+
|
29 |
+
for i in range(len(history)):
|
30 |
+
prompt+=history[i][0]+'\nASSISTANT: '+history[i][1]+"\nUSER: "
|
31 |
+
|
32 |
+
prompt = prompt+new_message+'\nASSISTANT: '
|
33 |
+
|
34 |
+
outputs = pipe(image, prompt=prompt, generate_kwargs={"max_new_tokens": 200
|
35 |
+
#, "do_sample" : True,
|
36 |
+
#"temperature" : 0.7
|
37 |
+
})[0]['generated_text']
|
38 |
+
|
39 |
+
return outputs[len(prompt)-6:]
|
40 |
+
|
41 |
+
import gradio as gr
|
42 |
+
|
43 |
+
with gr.Blocks() as demo:
|
44 |
+
|
45 |
+
with gr.Row():
|
46 |
+
image = gr.Image(type='pil', interactive=True)
|
47 |
+
|
48 |
+
gr.ChatInterface(
|
49 |
+
update_conversation, additional_inputs=[image]
|
50 |
+
)
|
51 |
+
|
52 |
+
demo.launch(debug=True)
|