Upload app.py with huggingface_hub

app.py

@@ -1,25 +1,45 @@
 """
-Mnemo v2 - Interactive Demo
+Mnemo v2 - Interactive Demo (ZeroGPU Compatible)
 Enhanced memory system with real embeddings, HNSW index, and temporal decay.
 """
 
 import gradio as gr
+import spaces
 import time
 from datetime import datetime
-from mnemo_core import get_mnemo
+from mnemo_core import get_mnemo, compute_embedding, compute_embeddings_batch
 
 def format_time(timestamp: float) -> str:
     return datetime.fromtimestamp(timestamp).strftime("%Y-%m-%d %H:%M:%S")
 
+def get_stats_text(user_id: str = "default") -> str:
+    mnemo = get_mnemo()
+    stats = mnemo.get_stats(user_id=user_id or "default")
+    
+    return f"""**System Stats**
+- Total memories: {stats['total_memories']}
+- User memories: {stats['user_memory_count']}
+- Total users: {stats['total_users']}
+- Adds: {stats['total_adds']} | Updates: {stats['total_updates']} | Deletes: {stats['total_deletes']}
+- Searches: {stats['total_searches']}
+- Decay half-life: {stats['decay_half_life_days']:.1f} days"""
+
+
+@spaces.GPU(duration=30)
 def add_memory(content: str, importance: float, tags: str, user_id: str):
+    """Add a memory - requires GPU for embedding computation."""
     if not content.strip():
         return "❌ Please enter content", get_stats_text(user_id)
     
     mnemo = get_mnemo()
     tags_list = [t.strip() for t in tags.split(",") if t.strip()] if tags else []
     
-    result = mnemo.add(
+    # Compute embedding (GPU operation)
+    embedding = compute_embedding(content)
+    
+    result = mnemo.add_with_embedding(
         content=content,
+        embedding=embedding,
         user_id=user_id or "default",
         importance=importance,
         tags=tags_list
@@ -30,15 +50,21 @@ def add_memory(content: str, importance: float, tags: str, user_id: str):
     
     return f"{icon} {result['message']} (ID: {result['id']})", get_stats_text(user_id)
 
+
+@spaces.GPU(duration=30)
 def search_memories(query: str, k: int, min_score: float, user_id: str):
+    """Search memories - requires GPU for embedding computation."""
     if not query.strip():
         return "❌ Please enter a search query"
     
     mnemo = get_mnemo()
     start = time.time()
     
-    results = mnemo.search(
-        query=query,
+    # Compute query embedding (GPU operation)
+    query_embedding = compute_embedding(query)
+    
+    results = mnemo.search_with_embedding(
+        query_embedding=query_embedding,
         user_id=user_id or "default",
         k=k,
         min_score=min_score
@@ -66,7 +92,9 @@ def search_memories(query: str, k: int, min_score: float, user_id: str):
     
     return output
 
+
 def list_memories_ui(user_id: str, limit: int):
+    """List memories - no GPU needed."""
     mnemo = get_mnemo()
     memories = mnemo.list_memories(user_id=user_id or "default", limit=limit)
     
@@ -85,7 +113,9 @@ def list_memories_ui(user_id: str, limit: int):
     
     return output
 
+
 def delete_memory_ui(memory_id: str, user_id: str):
+    """Delete a memory - no GPU needed."""
     if not memory_id.strip():
         return "❌ Please enter a memory ID", get_stats_text(user_id)
     
@@ -96,24 +126,17 @@ def delete_memory_ui(memory_id: str, user_id: str):
         return f"✅ Deleted memory: {memory_id}", get_stats_text(user_id)
     return f"❌ Memory not found: {memory_id}", get_stats_text(user_id)
 
+
 def clear_memories_ui(user_id: str):
+    """Clear all memories - no GPU needed."""
     mnemo = get_mnemo()
     count = mnemo.clear(user_id=user_id or "default")
     return f"🗑️ Cleared {count} memories", get_stats_text(user_id)
 
-def get_stats_text(user_id: str = "default") -> str:
-    mnemo = get_mnemo()
-    stats = mnemo.get_stats(user_id=user_id or "default")
-    
-    return f"""**System Stats**
-- Total memories: {stats['total_memories']}
-- User memories: {stats['user_memory_count']}
-- Total users: {stats['total_users']}
-- Adds: {stats['total_adds']} | Updates: {stats['total_updates']} | Deletes: {stats['total_deletes']}
-- Searches: {stats['total_searches']}
-- Decay half-life: {stats['decay_half_life_days']:.1f} days"""
 
+@spaces.GPU(duration=60)
 def load_examples(user_id: str):
+    """Load example memories - requires GPU for batch embedding computation."""
     mnemo = get_mnemo()
     
     examples = [
@@ -127,10 +150,15 @@ def load_examples(user_id: str):
         ("Lives in San Francisco, timezone is PST", 0.6, ["personal", "location"]),
     ]
     
+    # Batch compute embeddings for efficiency
+    texts = [ex[0] for ex in examples]
+    embeddings = compute_embeddings_batch(texts)
+    
     added = 0
-    for content, importance, tags in examples:
-        result = mnemo.add(
+    for i, (content, importance, tags) in enumerate(examples):
+        result = mnemo.add_with_embedding(
             content=content,
+            embedding=embeddings[i],
             user_id=user_id or "default",
             importance=importance,
             tags=tags
@@ -140,6 +168,7 @@ def load_examples(user_id: str):
     
     return f"✅ Loaded {added} example memories", get_stats_text(user_id)
 
+
 # Build the Gradio interface
 with gr.Blocks(title="Mnemo v2", theme=gr.themes.Soft()) as demo:
     gr.Markdown("""
@@ -152,6 +181,7 @@ with gr.Blocks(title="Mnemo v2", theme=gr.themes.Soft()) as demo:
     - 📊 Composite relevance scoring (similarity + recency + frequency)
     - 🔄 Intelligent ADD/UPDATE detection
     - 👥 Multi-user support
+    - 🎮 ZeroGPU powered (H200)
     """)
     
     with gr.Row():
@@ -253,6 +283,7 @@ with gr.Blocks(title="Mnemo v2", theme=gr.themes.Soft()) as demo:
     | Vector Index | `faiss-cpu` HNSW (M=32, efSearch=50) |
     | Decay | Ebbinghaus curve, 7-day half-life |
     | Scoring | 40% similarity + 30% recency + 30% frequency |
+    | GPU | ZeroGPU (H200, dynamic allocation) |
     
     **Links:** [GitHub](https://huggingface.co/AthelaPerk/mnemo-memory) | [API](https://huggingface.co/spaces/AthelaPerk/mnemo-mcp)
     """)