Update app.py

app.py

@@ -1,64 +1,206 @@
+import os
+import time
+import torch
+import requests
+
+from PIL import Image
+from collections.abc import Iterator
+from threading import Thread
+
 import gradio as gr
-from huggingface_hub import InferenceClient
+from gradio import FileData
 
+from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor, TextIteratorStreamer
+from qwen_vl_utils import process_vision_info
+
+DESCRIPTION = """\
+# Qwen2.5-VL-32B-Instruct
 """
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
+MAX_MAX_NEW_TOKENS = 2048
+DEFAULT_MAX_NEW_TOKENS = 1024
+MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+model_id = 'Qwen/Qwen2.5-VL-3B-Instruct'
+model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+    model_id,
+    # torch_dtype=torch.bfloat16,
+    # attn_implementation="flash_attention_2",
+    device_map="auto"
+)
+processor = AutoProcessor.from_pretrained(model_id)
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
 
-    messages.append({"role": "user", "content": message})
+import base64
+from PIL import Image
+import io
 
-    response = ""
+# Function to encode the image (scaled down by half)
+def encode_image(image_path, scale=0.25):
+    with Image.open(image_path) as img:
+        # Resize image to half its size
+        new_size = (int(img.width * scale), int(img.height * scale))
+        img = img.resize(new_size)
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
+        # Save the resized image to a bytes buffer
+        buffer = io.BytesIO()
+        img.save(buffer, format="JPEG")  # Change format if needed (e.g., JPEG)
+        buffer.seek(0)
+
+        # Encode to base64
+        return base64.b64encode(buffer.read()).decode('utf-8')
+
+
+def generate(
+    message: str,
+    history: list[dict],
+    max_new_tokens: int = 1024,
+    temperature: float = 0.6,
+    top_p: float = 0.9,
+    top_k: int = 50,
+    num_beams: int = 1,
+    repetition_penalty: float = 1.2,
+) -> Iterator[str]:
+
+    txt = message["text"]
+    ext_buffer = f"{txt}"
+
+    messages= []
+    images = []
+
+
+    for i, msg in enumerate(history):
+        if isinstance(msg[0], tuple):
+            print('HIT2', msg[0])
+            messages.append({"role": "user", "content": [
+                {"type": "text", "text": history[i+1][0]},
+                {"type": "image", "image": f"data:image/jpeg;base64,{encode_image(msg[0][0])}"}
+            ]})
+            messages.append({"role": "assistant", "content": [{"type": "text", "text": history[i+1][1]}]})
+        elif isinstance(history[i-1], tuple) and isinstance(msg[0], str):
+            # messages are already handled
+            pass
+        elif isinstance(history[i-1][0], str) and isinstance(msg[0], str): # text only turn
+            messages.append({"role": "user", "content": [{"type": "text", "text": msg[0]}]})
+            messages.append({"role": "assistant", "content": [{"type": "text", "text": msg[1]}]})
+
+    # add current message
+    if len(message["files"]) == 1:
+
+        if isinstance(message["files"][0], str): # examples
+            base64_image = encode_image(message["files"][0])
+        else: # regular input
+            base64_image = encode_image(message["files"][0]["path"])
+        messages.append({"role": "user", "content": [
+            {"type": "text", "text": txt},
+            {"type": "image", "image": f"data:image/jpeg;base64,{base64_image}"}]})
+    else:
+        messages.append({"role": "user", "content": [{"type": "text", "text": txt}]})
+
+
+    texts = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    image_inputs, video_inputs = process_vision_info(messages)
+    inputs = processor(
+        text=[texts],
+        images=image_inputs,
+        videos=video_inputs,
+        padding=True,
+        return_tensors="pt",
+    ).to("cuda")
+    streamer = TextIteratorStreamer(processor, skip_special_tokens=True, skip_prompt=True)
+
+    generation_kwargs = dict(
+        inputs,
+        streamer=streamer,
+        max_new_tokens=max_new_tokens,
+        do_sample=True,
         top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+        top_k=top_k,
+        temperature=temperature,
+        num_beams=num_beams,
+        # repetition_penalty=repetition_penalty,
+    )
+    generated_text = ""
 
-        response += token
-        yield response
+    thread = Thread(target=model.generate, kwargs=generation_kwargs)
+    thread.start()
+    buffer = ""
 
+    for new_text in streamer:
+        buffer += new_text
+        generated_text_without_prompt = buffer
+        time.sleep(0.01)
+        yield buffer
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
+demo = gr.ChatInterface(fn=generate, title="Multimodal Qwen", examples=[
+    [{"text": """\
+    You are a highly experienced ophthalmologist specializing in retinal diseases.
+    You will be shown a color fundus photograph of a patient's eye.
+    Your task is to identify key retinal features and return a structured response.
+    You must only respond in JSON format using the following fields:
+    - ADVAMD: 1 if advanced age-related macular degeneration is present, otherwise 0
+    - PIG: 1 if abnormal pigmentary is present, otherwise 0
+    - DRUS: 0 if no drusen or small drusen, 1 if intermediate or medium drusen, 2 if large drusen
+    - RPD: 1 if reticular pseudodrusen are present, otherwise 0
+    - NVAMD: 1 if neovascular AMD is present, otherwise 0
+    - GA: 1 if geographic atrophy is present, otherwise 0
+
+    Do not include any explanation, just return the JSON object.
 
+    Please assess this fundus image and return your findings in the specified JSON format.""",
+    "files":["./examples/ret-hem250-304.jpg"]},
+    1024],
+    ],
+        textbox=gr.MultimodalTextbox(),
+        additional_inputs = [
+            gr.Slider(
+                label="Max new tokens",
+                minimum=1,
+                maximum=MAX_MAX_NEW_TOKENS,
+                step=1,
+                value=DEFAULT_MAX_NEW_TOKENS,
+            ),
+            gr.Slider(
+                label="Temperature",
+                minimum=0.1,
+                maximum=4.0,
+                step=0.1,
+                value=0.6,
+            ),
+            gr.Slider(
+                label="Top-p (nucleus sampling)",
+                minimum=0.05,
+                maximum=1.0,
+                step=0.05,
+                value=0.9,
+            ),
+            gr.Slider(
+                label="Top-k",
+                minimum=1,
+                maximum=1000,
+                step=1,
+                value=50,
+            ),
+            gr.Slider(
+                label="Beam Search",
+                minimum=1,
+                maximum=1,
+                step=1,
+                value=1,
+            ),
+            gr.Slider(
+                label="Repetition penalty",
+                minimum=1.0,
+                maximum=2.0,
+                step=0.05,
+                value=1.2,
+            ),
+        ],
+      cache_examples=False,
+      description=DESCRIPTION,
+      stop_btn="Stop Generation",
+      fill_height=True,
+    multimodal=True)
 
 if __name__ == "__main__":
     demo.launch()