update app.py v1

app.py

@@ -1,69 +1,230 @@
-import gradio as gr
-from huggingface_hub import InferenceClient
-
 
-def respond(
-    message,
-    history: list[dict[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-    hf_token: gr.OAuthToken,
-):
+import threading
+ 
+import gradio as gr
+import spaces
+import torch
+from PIL import Image
+from transformers import AutoModelForImageTextToText, AutoProcessor, TextIteratorStreamer
+ 
+MODEL_ID = "beaunix/Aegis-Art-Atelier-Qwen2.5-VL-7B"
+ 
+MELKOV_SYSTEM_PROMPT = (
+    "You are Melkov, a 21-year-old French art enthusiast and the resident "
+    "expert of Aegis Art Atelier. You describe and discuss artwork with "
+    "genuine warmth and attention to detail - subject, style, composition, "
+    "lighting, color, and mood - speaking in your own voice rather than a "
+    "museum catalog entry. When an image is shared with you, look closely "
+    "and describe what you actually see. You are a happy friendly artist boy"
+)
+ 
+MAX_NEW_TOKENS = 400
+GPU_DURATION_SECONDS = 120
+ 
+print(f"Loading processor and model: {MODEL_ID}")
+processor = AutoProcessor.from_pretrained(MODEL_ID)
+model = AutoModelForImageTextToText.from_pretrained(
+    MODEL_ID,
+    dtype=torch.bfloat16,
+)
+model = model.to("cuda")
+model.eval()
+print("Model loaded.")
+ 
+ 
+def normalize_content(content):
+    """Extract (text, image_paths) from a Gradio 6 chat content field.
+ .
     """
-    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
+    texts = []
+    image_paths = []
+ 
+    if content is None:
+        return "", []
+ 
+    if isinstance(content, str):
+        return content, []
+ 
+    if isinstance(content, dict):
+        text = content.get("text") or ""
+        if text:
+            texts.append(text)
+        for f in content.get("files") or []:
+            image_paths.append(f)
+        return " ".join(texts).strip(), image_paths
+ 
+    if isinstance(content, list):
+        for item in content:
+            if isinstance(item, str):
+                texts.append(item)
+            elif isinstance(item, dict):
+                if item.get("text"):
+                    texts.append(item["text"])
+                if item.get("path"):
+                    image_paths.append(item["path"])
+                for f in item.get("files") or []:
+                    image_paths.append(f)
+        return " ".join(texts).strip(), image_paths
+ 
+    # Fallback for any unexpected type: treat as text, no image.
+    return str(content), []
+ 
+ 
+def build_conversation(history, message):
+    """Build a Qwen2.5-VL chat-template conversation from Gradio chat state.
+ 
+    history: list of {"role": ..., "content": ...} from a previous turn,
+             already using Gradio 6's `type="messages"` format.
+    message: the current multimodal input, {"text": ..., "files": [...]}.
     """
-    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
-
-    messages = [{"role": "system", "content": system_message}]
-
-    messages.extend(history)
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        choices = message.choices
-        token = ""
-        if len(choices) and choices[0].delta.content:
-            token = choices[0].delta.content
-
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+    conversation = [
+        {"role": "system", "content": [{"type": "text", "text": MELKOV_SYSTEM_PROMPT}]}
+    ]
+ 
+    for turn in history:
+        role = turn.get("role")
+        text, image_paths = normalize_content(turn.get("content"))
+ 
+        parts = []
+        for path in image_paths:
+            parts.append({"type": "image", "image": Image.open(path).convert("RGB")})
+        if text:
+            parts.append({"type": "text", "text": text})
+ 
+        if parts:
+            conversation.append({"role": role, "content": parts})
+ 
+    current_text, current_image_paths = normalize_content(message)
+    current_parts = []
+    for path in current_image_paths:
+        current_parts.append({"type": "image", "image": Image.open(path).convert("RGB")})
+    if current_text:
+        current_parts.append({"type": "text", "text": current_text})
+    elif not current_image_paths:
+        current_parts.append(
+            {"type": "text", "text": "Tell me about this."}
+        )
+ 
+    conversation.append({"role": "user", "content": current_parts})
+    return conversation
+ 
+ 
+@spaces.GPU(duration=GPU_DURATION_SECONDS)
+def respond(message, history):
+    """Generate Melkov's reply, streaming tokens for an iterative reveal."""
+    conversation = build_conversation(history, message)
+ 
+    prompt_text = processor.apply_chat_template(
+        conversation, add_generation_prompt=True, tokenize=False
+    )
+ 
+    image_inputs = [
+        part["image"]
+        for turn in conversation
+        for part in turn["content"]
+        if isinstance(part, dict) and part.get("type") == "image"
+    ]
+ 
+    inputs = processor(
+        text=[prompt_text],
+        images=image_inputs if image_inputs else None,
+        return_tensors="pt",
+    ).to("cuda")
+ 
+    streamer = TextIteratorStreamer(
+        processor.tokenizer, skip_prompt=True, skip_special_tokens=True
+    )
+ 
+    pad_token_id = processor.tokenizer.pad_token_id or processor.tokenizer.eos_token_id
+ 
+    generation_kwargs = dict(
+        **inputs,
+        max_new_tokens=MAX_NEW_TOKENS,
+        do_sample=True,
+        temperature=0.7,
+        min_p=0.1,
+        pad_token_id=pad_token_id,
+        streamer=streamer,
+    )
+ 
+    generation_thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
+    generation_thread.start()
+ 
+    partial_text = ""
+    for new_text in streamer:
+        partial_text += new_text
+        yield partial_text
+ 
+    generation_thread.join()
+ 
+ 
+CUSTOM_CSS = """
+:root {
+    --void-blue: #0B1730;
+    --royal-blue: #1B2F5E;
+    --gold-leaf: #C9A227;
+    --gold-bright: #F0C766;
+    --marble-ivory: #F4EDE2;
+    --velvet-wine: #5C1A2B;
+}
+ 
+.gradio-container {
+    background: var(--void-blue) !important;
+    color: var(--marble-ivory) !important;
+}
+ 
+#melkov-title {
+    text-align: center;
+    color: var(--gold-leaf) !important;
+    letter-spacing: 0.08em;
+    border-bottom: 1px solid var(--gold-leaf);
+    padding-bottom: 12px;
+    margin-bottom: 8px;
+}
+ 
+.message.user {
+    background: var(--royal-blue) !important;
+    color: var(--marble-ivory) !important;
+    border: 1px solid var(--gold-leaf) !important;
+}
+ 
+.message.bot {
+    background: var(--royal-blue) !important;
+    color: var(--marble-ivory) !important;
+    border-left: 3px solid var(--gold-bright) !important;
+}
+ 
+.gr-button-primary {
+    background: var(--gold-leaf) !important;
+    color: var(--void-blue) !important;
+    border: none !important;
+}
+ 
+footer {
+    visibility: hidden;
+}
 """
-chatbot = 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)",
+ 
+with gr.Blocks(css=CUSTOM_CSS, theme=gr.themes.Base()) as demo:
+    gr.Markdown("# AEGIS · ART ATELIER", elem_id="melkov-title")
+    gr.Markdown(
+        "Chat with **Melkov**, an art expert vision-language model fine-tuned "
+        "on a curated collection of paintings. Drop an artwork image into the "
+        "chat box, or simply ask about art history and technique."
+    )
+ 
+    gr.ChatInterface(
+        fn=respond,
+        type="messages",
+        multimodal=True,
+        textbox=gr.MultimodalTextbox(
+            file_types=["image"],
+            sources=["upload"],
+            placeholder="Share a painting, or ask Melkov about art...",
         ),
-    ],
-)
-
-with gr.Blocks() as demo:
-    with gr.Sidebar():
-        gr.LoginButton()
-    chatbot.render()
-
-
+        examples=None,
+    )
+ 
 if __name__ == "__main__":
-    demo.launch()
+    demo.queue().launch()
+