Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from gradio.data_classes import FileData
|
3 |
+
from huggingface_hub import snapshot_download
|
4 |
+
from pathlib import Path
|
5 |
+
import base64
|
6 |
+
import spaces
|
7 |
+
import os
|
8 |
+
|
9 |
+
from mistral_inference.transformer import Transformer
|
10 |
+
from mistral_inference.generate import generate
|
11 |
+
|
12 |
+
from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
|
13 |
+
from mistral_common.protocol.instruct.messages import UserMessage, TextChunk, ImageURLChunk
|
14 |
+
from mistral_common.protocol.instruct.request import ChatCompletionRequest
|
15 |
+
|
16 |
+
models_path = Path.home().joinpath('pixtral', 'Pixtral')
|
17 |
+
models_path.mkdir(parents=True, exist_ok=True)
|
18 |
+
|
19 |
+
snapshot_download(repo_id="mistral-community/pixtral-12b-240910",
|
20 |
+
allow_patterns=["params.json", "consolidated.safetensors", "tekken.json"],
|
21 |
+
local_dir=models_path)
|
22 |
+
|
23 |
+
tokenizer = MistralTokenizer.from_file(f"{models_path}/tekken.json")
|
24 |
+
model = Transformer.from_folder(models_path)
|
25 |
+
|
26 |
+
def image_to_base64(image_path):
|
27 |
+
with open(image_path, 'rb') as img:
|
28 |
+
encoded_string = base64.b64encode(img.read()).decode('utf-8')
|
29 |
+
return f"data:image/jpeg;base64,{encoded_string}"
|
30 |
+
|
31 |
+
@spaces.GPU
|
32 |
+
def run_inference(image_url, prompt):
|
33 |
+
base64 = image_to_base64(image_url)
|
34 |
+
completion_request = ChatCompletionRequest(messages=[UserMessage(content=[ImageURLChunk(image_url=base64), TextChunk(text=prompt)])])
|
35 |
+
|
36 |
+
encoded = tokenizer.encode_chat_completion(completion_request)
|
37 |
+
|
38 |
+
images = encoded.images
|
39 |
+
tokens = encoded.tokens
|
40 |
+
|
41 |
+
out_tokens, _ = generate([tokens], model, images=[images], max_tokens=512, temperature=0.45, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
|
42 |
+
result = tokenizer.decode(out_tokens[0])
|
43 |
+
return [[prompt, result]]
|
44 |
+
|
45 |
+
with gr.Blocks() as demo:
|
46 |
+
with gr.Row():
|
47 |
+
image_box = gr.Image(type="filepath")
|
48 |
+
|
49 |
+
chatbot = gr.Chatbot(
|
50 |
+
scale = 2,
|
51 |
+
height=750
|
52 |
+
)
|
53 |
+
text_box = gr.Textbox(
|
54 |
+
placeholder="Enter text and press enter, or upload an image",
|
55 |
+
container=False,
|
56 |
+
)
|
57 |
+
|
58 |
+
|
59 |
+
btn = gr.Button("Submit")
|
60 |
+
clicked = btn.click(run_inference,
|
61 |
+
[image_box,text_box],
|
62 |
+
chatbot
|
63 |
+
)
|
64 |
+
|
65 |
+
demo.queue().launch()
|