Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import spaces
|
| 2 |
+
import gradio as gr
|
| 3 |
+
from transformers import pipeline, TextIteratorStreamer
|
| 4 |
+
import torch
|
| 5 |
+
import threading
|
| 6 |
+
|
| 7 |
+
# Load model and tokenizer
|
| 8 |
+
model_name = "krish10/Qwen3_14B_16bit_TA_screen_v1.0"
|
| 9 |
+
pipe = pipeline("text-generation", model=model_name, device=0)
|
| 10 |
+
tokenizer = pipe.tokenizer
|
| 11 |
+
model = pipe.model
|
| 12 |
+
|
| 13 |
+
# Fixed generation config
|
| 14 |
+
MAX_TOKENS = 3000
|
| 15 |
+
TEMPERATURE = 0.1
|
| 16 |
+
TOP_P = 0.9
|
| 17 |
+
|
| 18 |
+
@spaces.GPU
|
| 19 |
+
def respond_stream(title, abstract):
|
| 20 |
+
if not title.strip() or not abstract.strip():
|
| 21 |
+
return "❌ Error: Title and Abstract are required."
|
| 22 |
+
|
| 23 |
+
prompt = f"Title: {title.strip()}\nAbstract: {abstract.strip()}"
|
| 24 |
+
|
| 25 |
+
messages = [{"role": "user", "content": prompt}]
|
| 26 |
+
prompt_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
| 27 |
+
|
| 28 |
+
inputs = tokenizer(prompt_text, return_tensors="pt").to("cuda")
|
| 29 |
+
streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
|
| 30 |
+
|
| 31 |
+
generation_kwargs = dict(
|
| 32 |
+
input_ids=inputs["input_ids"],
|
| 33 |
+
streamer=streamer,
|
| 34 |
+
max_new_tokens=MAX_TOKENS,
|
| 35 |
+
temperature=TEMPERATURE,
|
| 36 |
+
top_p=TOP_P,
|
| 37 |
+
do_sample=True,
|
| 38 |
+
pad_token_id=tokenizer.eos_token_id,
|
| 39 |
+
)
|
| 40 |
+
|
| 41 |
+
thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
|
| 42 |
+
thread.start()
|
| 43 |
+
|
| 44 |
+
partial_text = ""
|
| 45 |
+
for token in streamer:
|
| 46 |
+
partial_text += token
|
| 47 |
+
yield partial_text
|
| 48 |
+
|
| 49 |
+
# Build Gradio interface
|
| 50 |
+
with gr.Blocks() as demo:
|
| 51 |
+
gr.Markdown("## 🤖 Qwen Streaming Chat — Medical Abstract Evaluator")
|
| 52 |
+
|
| 53 |
+
with gr.Column():
|
| 54 |
+
title = gr.Textbox(label="Title", lines=2, placeholder="Required")
|
| 55 |
+
abstract = gr.Textbox(label="Abstract", lines=10, placeholder="Required")
|
| 56 |
+
|
| 57 |
+
output_box = gr.Textbox(label="Model Response", lines=15, interactive=False)
|
| 58 |
+
generate_btn = gr.Button("Generate")
|
| 59 |
+
|
| 60 |
+
generate_btn.click(
|
| 61 |
+
fn=respond_stream,
|
| 62 |
+
inputs=[title, abstract],
|
| 63 |
+
outputs=[output_box]
|
| 64 |
+
)
|
| 65 |
+
|
| 66 |
+
# Launch the app
|
| 67 |
+
if __name__ == "__main__":
|
| 68 |
+
demo.launch()
|