Update app.py

app.py

@@ -1,64 +1,92 @@
-import gradio as gr
-from huggingface_hub import InferenceClient
-
 """
-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
+PetBull-7B-VL demo Space
+------------------------
+
+• Base model  : Qwen/Qwen2.5-VL-7B-Instruct
+• LoRA adapter: ColdSlim/PetBull-7B   (light-weight repo you just pushed)
+
+Put this file in your Space, add a `requirements.txt` with:
+    transformers>=4.41.0
+    peft>=0.11.0
+    accelerate
+    gradio>=4.33
+
+Then (optionally) switch the Space hardware to **GPU (shared)** in
+Settings → Hardware for much faster vision-language inference.
 """
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
+import torch, gradio as gr
+from PIL import Image
+from transformers import AutoProcessor, AutoModelForVision2Seq
+from peft import PeftModel
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+# ---------------------------------------------------------------------
+# 1  Load base + LoRA  (≈ 12 GB VRAM in bf16; falls back to CPU if needed)
+# ---------------------------------------------------------------------
+BASE_MODEL   = "Qwen/Qwen2.5-VL-7B-Instruct"
+ADAPTER_REPO = "ColdSlim/PetBull-7B"     # 👉 replace with your HF path if different
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+device = "cuda" if torch.cuda.is_available() else "cpu"
+dtype  = torch.float16 if device == "cuda" else torch.float32
 
-    messages.append({"role": "user", "content": message})
+processor = AutoProcessor.from_pretrained(BASE_MODEL, trust_remote_code=True)
+base      = AutoModelForVision2Seq.from_pretrained(
+                BASE_MODEL, torch_dtype=dtype, trust_remote_code=True)
+model     = PeftModel.from_pretrained(base, ADAPTER_REPO)
+model.to(device).eval()
 
-    response = ""
+# ---------------------------------------------------------------------
+# 2  Inference helper
+# ---------------------------------------------------------------------
+def generate_answer(image: Image.Image | None,
+                    question: str,
+                    temperature: float = 0.7,
+                    top_p: float = 0.95,
+                    max_tokens: int = 512) -> str:
+    """
+    Runs one‐shot VQA chat. Image is optional; if None we still obey the
+    prompt format required by Qwen-VL by inserting a blank white image.
+    """
+    if image is None:
+        image = Image.new("RGB", (224, 224), color="white")
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+    inputs = processor(text=[question],
+                       images=[image],
+                       return_tensors="pt").to(device)
 
-        response += token
-        yield response
+    with torch.no_grad():
+        output_ids = model.generate(**inputs,
+                                    max_new_tokens=max_tokens,
+                                    temperature=temperature,
+                                    top_p=top_p)
+    return processor.batch_decode(output_ids,
+                                  skip_special_tokens=True)[0]
 
+# ---------------------------------------------------------------------
+# 3  Gradio UI
+# ---------------------------------------------------------------------
+with gr.Blocks(title="PetBull-7B-VL – Ask a Vet Bot") as demo:
+    gr.Markdown(
+        """
+        ## 🐾 PetBull-7B-VL  
+        Upload a photo of your pet **and/or** ask a question.  
+        The model will analyse the image (if provided) and give tailored advice.
+        """
+    )
 
-"""
-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)",
-        ),
-    ],
-)
+    with gr.Row():
+        with gr.Column(scale=1):
+            img_in   = gr.Image(type="pil", label="Pet photo (optional)")
+            txt_in   = gr.Textbox(lines=3, placeholder="Describe the issue or ask a question…")
+            run_btn  = gr.Button("Ask PetBull")
+            temp_sl  = gr.Slider(0.1, 1.5, 0.7, label="Temperature")
+            topp_sl  = gr.Slider(0.1, 1.0, 0.95, label="Top-p")
+            max_sl   = gr.Slider(32, 1024, 512, step=8, label="Max new tokens")
+        with gr.Column(scale=1):
+            answer   = gr.Textbox(lines=12, label="Assistant", interactive=False)
 
+    run_btn.click(fn=generate_answer,
+                  inputs=[img_in, txt_in, temp_sl, topp_sl, max_sl],
+                  outputs=answer)
 
-if __name__ == "__main__":
-    demo.launch()
+demo.queue().launch()