Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,97 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from __future__ import annotations
|
2 |
+
|
3 |
+
import os
|
4 |
+
import hashlib
|
5 |
+
import torch
|
6 |
+
from threading import Thread
|
7 |
+
from transformers import AutoModel, AutoProcessor, TextIteratorStreamer
|
8 |
+
import gradio as gr
|
9 |
+
|
10 |
+
# Initialize the model and processor
|
11 |
+
def initialize_model_and_processor():
|
12 |
+
model = AutoModel.from_pretrained("OEvortex/HelpingAI-Vision", torch_dtype=torch.float16, trust_remote_code=True).to("cuda" if torch.cuda.is_available() else "cpu")
|
13 |
+
processor = AutoProcessor.from_pretrained("OEvortex/HelpingAI-Vision", trust_remote_code=True)
|
14 |
+
return model, processor
|
15 |
+
|
16 |
+
# Function to process images and cache results
|
17 |
+
def cached_vision_process(image, max_crops, num_tokens):
|
18 |
+
image_hash = hashlib.sha256(image.tobytes()).hexdigest()
|
19 |
+
cache_path = f"visual_cache/{image_hash}-{max_crops}-{num_tokens}.pt"
|
20 |
+
if os.path.exists(cache_path):
|
21 |
+
return torch.load(cache_path).to(model.device, dtype=model.dtype)
|
22 |
+
else:
|
23 |
+
processor_outputs = processor.image_processor([image], max_crops)
|
24 |
+
pixel_values = [value.to(model.device, model.dtype) for value in processor_outputs["pixel_values"]]
|
25 |
+
coords = [value.to(model.device, model.dtype) for value in processor_outputs["coords"]]
|
26 |
+
image_outputs = model.vision_model(pixel_values, coords, num_tokens)
|
27 |
+
image_features = model.multi_modal_projector(image_outputs)
|
28 |
+
os.makedirs("visual_cache", exist_ok=True)
|
29 |
+
torch.save(image_features, cache_path)
|
30 |
+
return image_features.to(model.device, model.dtype)
|
31 |
+
|
32 |
+
# Function to answer questions about images
|
33 |
+
def answer_question(image, question, max_crops, num_tokens, sample, temperature, top_k):
|
34 |
+
if not question.strip() or not image:
|
35 |
+
return "Please provide both an image and a question."
|
36 |
+
|
37 |
+
prompt = f"""user
|
38 |
+
<image>
|
39 |
+
{question}
|
40 |
+
assistant
|
41 |
+
"""
|
42 |
+
streamer = TextIteratorStreamer(processor.tokenizer, skip_special_tokens=True)
|
43 |
+
with torch.inference_mode():
|
44 |
+
inputs = processor(prompt, [image], model, max_crops=max_crops, num_tokens=num_tokens)
|
45 |
+
|
46 |
+
generation_kwargs = {
|
47 |
+
"input_ids": inputs["input_ids"],
|
48 |
+
"attention_mask": inputs["attention_mask"],
|
49 |
+
"image_features": inputs["image_features"],
|
50 |
+
"streamer": streamer,
|
51 |
+
"max_length": 1000,
|
52 |
+
"use_cache": True,
|
53 |
+
"eos_token_id": processor.tokenizer.eos_token_id,
|
54 |
+
"pad_token_id": processor.tokenizer.eos_token_id,
|
55 |
+
"temperature": temperature,
|
56 |
+
"do_sample": sample,
|
57 |
+
"top_k": top_k,
|
58 |
+
}
|
59 |
+
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
60 |
+
thread.start()
|
61 |
+
|
62 |
+
buffer = ""
|
63 |
+
output_started = False
|
64 |
+
for new_text in streamer:
|
65 |
+
if not output_started:
|
66 |
+
if "assistant" in new_text:
|
67 |
+
output_started = True
|
68 |
+
continue
|
69 |
+
buffer += new_text
|
70 |
+
if len(buffer) > 1:
|
71 |
+
yield buffer
|
72 |
+
return buffer
|
73 |
+
|
74 |
+
# Initialize the model and processor
|
75 |
+
model, processor = initialize_model_and_processor()
|
76 |
+
|
77 |
+
# Gradio interface setup
|
78 |
+
with gr.Blocks() as demo:
|
79 |
+
with gr.Group():
|
80 |
+
with gr.Row():
|
81 |
+
prompt = gr.Textbox(label="Question", placeholder="e.g. Describe this?", scale=4)
|
82 |
+
submit = gr.Button("Send", scale=1)
|
83 |
+
with gr.Row():
|
84 |
+
max_crops = gr.Slider(minimum=0, maximum=200, step=5, value=0, label="Max crops")
|
85 |
+
num_tokens = gr.Slider(minimum=728, maximum=2184, step=10, value=728, label="Number of image tokens")
|
86 |
+
with gr.Row():
|
87 |
+
img = gr.Image(type="pil", label="Upload or Drag an Image")
|
88 |
+
output = gr.TextArea(label="Answer")
|
89 |
+
with gr.Row():
|
90 |
+
sample = gr.Checkbox(label="Sample", value=False)
|
91 |
+
temperature = gr.Slider(minimum=0, maximum=1, step=0.1, value=0, label="Temperature")
|
92 |
+
top_k = gr.Slider(minimum=0, maximum=50, step=1, value=0, label="Top-K")
|
93 |
+
|
94 |
+
submit.click(answer_question, [img, prompt, max_crops, num_tokens, sample, temperature, top_k], output)
|
95 |
+
prompt.submit(answer_question, [img, prompt, max_crops, num_tokens, sample, temperature, top_k], output)
|
96 |
+
|
97 |
+
demo.queue().launch(debug=True)
|