Upload folder using huggingface_hub

README.md

@@ -2,7 +2,7 @@
 title: WOOF
 emoji: "\U0001F415"
 colorFrom: yellow
-colorTo: yellow
+colorTo: green
 sdk: gradio
 sdk_version: "5.34.2"
 app_file: app.py
@@ -46,9 +46,9 @@ Same dad. Daughter has dyslexia. Mom has dementia. Dog has opinions about grass.
 
 ## The Stack
 
-- **Model:** MiniCPM5-1B (OpenBMB) via llama.cpp — 1B parameters
+- **Model:** NVIDIA Nemotron 3 Nano Omni via NIM API — multimodal (text + vision)
 - **Framework:** Gradio
-- **Cloud APIs:** Zero. Your dog's medical history stays private.
+- **Fallback:** Rule-based analysis when API is unavailable
 
 ## Important
 

app.py

@@ -13,31 +13,24 @@ their stomach hurts but their output can.
 """
 
 import gradio as gr
-# from llama_cpp import Llama
-from llama_cpp.llama_chat_format import MiniCPMv26ChatHandler
-
-# --- Model ---
-# MiniCPM-V for vision (can analyze images) — OpenBMB sponsor model
-# Falls back to text-only analysis if vision model unavailable
-
-MODEL_REPO = "openbmb/MiniCPM5-1B-128k-GGUF"
-MODEL_FILE = "minicpm5-1b-128k-q4_k_m.gguf"
-
-print("Loading model...")
-try:
-    llm = Llama.from_pretrained(
-        repo_id=MODEL_REPO,
-        filename=MODEL_FILE,
-        n_ctx=2048,
-        n_threads=4,
-        verbose=False,
+import os
+import io
+import base64
+from openai import OpenAI
+
+# --- NVIDIA Nemotron Nano Omni (sponsor model) ---
+NVIDIA_API_KEY = os.environ.get("NVIDIA_API_KEY", "")
+NEMOTRON_MODEL = "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning"
+
+nvidia_client = None
+if NVIDIA_API_KEY:
+    nvidia_client = OpenAI(
+        base_url="https://integrate.api.nvidia.com/v1",
+        api_key=NVIDIA_API_KEY,
     )
-    MODEL_LOADED = False
-    print("Model loaded.")
-except Exception as e:
-    print(f"Model failed to load: {e}")
-    MODEL_LOADED = False
-    llm = None
+    print("NVIDIA Nemotron connected.")
+else:
+    print("No NVIDIA_API_KEY — running in fallback mode.")
 
 
 # --- Dog Health Knowledge Base ---
@@ -115,7 +108,7 @@ FUN FACT: [one actually interesting dog gut fact to lighten the mood]"""
 def analyze_output(description, output_type, dog_size, photo):
     """Analyze dog output from description and/or photo."""
     if not description and not photo:
-        return """<div class="wf-empty">Describe what you found or upload a photo. No judgment here — we've all been there at 6am.</div>"""
+        return """<div class="wf-empty">Describe what you found or upload a photo. No judgment here.</div>"""
 
     # Check for known toxic items in description
     alerts = []
@@ -133,28 +126,41 @@ def analyze_output(description, output_type, dog_size, photo):
             if key in lower:
                 color_match = info
 
-    # Build context
+    # Build the prompt
     context_parts = []
     if output_type: context_parts.append(f"Type: {output_type}")
     if dog_size: context_parts.append(f"Dog size: {dog_size}")
     context = " | ".join(context_parts)
+    user_text = f"My dog's {output_type or 'poop'}: {description or 'see photo'}"
+    if context: user_text += f"\n({context})"
 
-    if MODEL_LOADED and llm:
+    # Try Nemotron
+    if nvidia_client:
         try:
-            prompt = f"My dog's {output_type or 'poop'}: {description or 'see photo'}"
-            if context: prompt += f"\n({context})"
-
-            response = llm.create_chat_completion(
+            content = [{"type": "text", "text": user_text}]
+
+            if photo:
+                buffered = io.BytesIO()
+                photo.save(buffered, format="JPEG", quality=85)
+                b64 = base64.b64encode(buffered.getvalue()).decode()
+                content.append({
+                    "type": "image_url",
+                    "image_url": {"url": f"data:image/jpeg;base64,{b64}"}
+                })
+
+            response = nvidia_client.chat.completions.create(
+                model=NEMOTRON_MODEL,
                 messages=[
                     {"role": "system", "content": SYSTEM_PROMPT},
-                    {"role": "user", "content": prompt}
+                    {"role": "user", "content": content}
                 ],
                 max_tokens=600,
                 temperature=0.3,
             )
-            raw = response["choices"][0]["message"]["content"].strip()
+            raw = response.choices[0].message.content.strip()
             return format_analysis(raw, alerts, color_match)
-        except:
+        except Exception as e:
+            print(f"Nemotron error: {e}")
             return format_analysis(fallback_analysis(description, output_type), alerts, color_match)
     else:
         return format_analysis(fallback_analysis(description, output_type), alerts, color_match)
@@ -430,7 +436,7 @@ with gr.Blocks(css=CUSTOM_CSS, title="WOOF", theme=gr.themes.Soft()) as app:
     gr.HTML("""
     <div style="text-align:center; padding: 16px 0 4px; color: #CCC; font-size: 0.7em; font-family: 'Quicksand', sans-serif; line-height: 1.8;">
         Not a substitute for veterinary care. When in doubt, call your vet.<br>
-        1B parameters. No internet needed. Your dog's business stays private.<br>
+        Powered by NVIDIA Nemotron Nano Omni<br>
         <span style="color:#8B6914;">Heuremen — Build Small Hackathon 2026</span>
     </div>
     """)

requirements.txt

@@ -1,3 +1,3 @@
 gradio>=4.0
-# llama-cpp-python
+openai
 Pillow