Create app.py

app.py

@@ -0,0 +1,153 @@
+import os
+import json
+import re
+
+os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
+os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True,max_split_size_mb:128"
+os.environ["HF_HOME"] = "/tmp/hf"
+os.environ["HF_HUB_CACHE"] = "/tmp/hf/hub"
+os.environ["TRANSFORMERS_CACHE"] = "/tmp/hf/transformers"
+
+import spaces
+import torch
+import gradio as gr
+from PIL import Image
+from transformers import AutoProcessor, AutoModelForImageTextToText
+
+MODEL_ID = "Qwen/Qwen3.5-397B-A17B"
+
+processor = None
+model = None
+
+
+def load_model():
+    global processor, model
+    if model is not None and processor is not None:
+        return
+
+    processor = AutoProcessor.from_pretrained(MODEL_ID)
+    model = AutoModelForImageTextToText.from_pretrained(
+        MODEL_ID,
+        device_map="auto",
+        torch_dtype="auto",
+    )
+    model.eval()
+
+
+def extract_json(text: str):
+    text = (text or "").strip()
+
+    try:
+        return json.loads(text)
+    except Exception:
+        pass
+
+    match = re.search(r"\{.*\}", text, flags=re.S)
+    if match:
+        try:
+            return json.loads(match.group(0))
+        except Exception:
+            pass
+
+    return {"raw_output": text}
+
+
+PROMPT = """Analyze this pantry image.
+Return ONLY valid JSON with this schema:
+{
+  "items": [
+    {
+      "name": "",
+      "brand": "",
+      "category": "",
+      "package_type": "",
+      "estimated_quantity": "",
+      "evidence": "",
+      "confidence": 0.0
+    }
+  ],
+  "summary": "",
+  "uncertain_items": []
+}
+
+Rules:
+- List visible pantry foods, ingredients, drinks, and packaged items.
+- Use the smallest sensible item name.
+- Do not invent hidden ingredients.
+- If a brand is unclear, leave brand empty.
+- If uncertain, lower confidence.
+- Do not include markdown, code fences, or commentary.
+"""
+
+
+@spaces.GPU(size="large", duration=60)
+def analyze_pantry(image: Image.Image):
+    if image is None:
+        return {"error": "Please upload a pantry image."}
+
+    load_model()
+
+    messages = [
+        {
+            "role": "system",
+            "content": [
+                {"type": "text", "text": "You extract pantry items from photos and respond with JSON only."}
+            ],
+        },
+        {
+            "role": "user",
+            "content": [
+                {"type": "image", "image": image.convert("RGB")},
+                {"type": "text", "text": PROMPT},
+            ],
+        },
+    ]
+
+    inputs = processor.apply_chat_template(
+        messages,
+        add_generation_prompt=True,
+        tokenize=True,
+        return_dict=True,
+        return_tensors="pt",
+    )
+
+    inputs = {k: v.to(model.device) if hasattr(v, "to") else v for k, v in inputs.items()}
+
+    with torch.inference_mode():
+        output_ids = model.generate(
+            **inputs,
+            max_new_tokens=1200,
+            do_sample=False,
+        )
+
+    prompt_len = inputs["input_ids"].shape[-1]
+    generated_text = processor.decode(
+        output_ids[0][prompt_len:],
+        skip_special_tokens=True,
+    ).strip()
+
+    parsed = extract_json(generated_text)
+    if isinstance(parsed, dict) and "raw_output" not in parsed:
+        parsed["_raw_output"] = generated_text
+    return parsed
+
+
+@spaces.GPU(size="large", duration=1)
+def cloud():
+    return None
+
+
+with gr.Blocks() as demo:
+    gr.Markdown("# Pantry ingredient / item extractor")
+
+    image_input = gr.Image(type="pil", label="Pantry image")
+    analyze_btn = gr.Button("Analyze")
+    cloud_btn = gr.Button("Cloud")
+    output_json = gr.JSON(label="Output")
+
+    analyze_btn.click(analyze_pantry, inputs=[image_input], outputs=[output_json], api_name="analyze")
+    cloud_btn.click(cloud, inputs=[], outputs=[], api_name="cloud")
+
+
+demo.queue(max_size=16)
+demo.launch(ssr_mode=False)