🔧 Switch to Inference API (no local model loading)

app.py

@@ -7,6 +7,40 @@ import torch
 # ============================================================================
 
 MODEL_ID = "OpceanAI/Yuuki-best"
+MODEL_LOADED = False
+model = None
+tokenizer = None
+
+
+def load_model():
+    """Load the Yuuki model with proper error handling."""
+    global model, tokenizer, MODEL_LOADED
+    
+    if MODEL_LOADED:
+        return True
+    
+    try:
+        print(f"Loading Yuuki model from {MODEL_ID}...")
+        
+        tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+        model = AutoModelForCausalLM.from_pretrained(
+            MODEL_ID,
+            torch_dtype=torch.float32,
+            low_cpu_mem_usage=True,
+            trust_remote_code=True
+        )
+        
+        if tokenizer.pad_token is None:
+            tokenizer.pad_token = tokenizer.eos_token
+        
+        MODEL_LOADED = True
+        print("Model loaded successfully!")
+        return True
+        
+    except Exception as e:
+        print(f"Error loading model: {e}")
+        return False
+
 
 def generate_code(
     prompt: str,
@@ -16,28 +50,43 @@ def generate_code(
     top_k: int = 50,
     repetition_penalty: float = 1.1
 ) -> str:
-    """Generate code using HuggingFace Inference API (no local loading)."""
+    """Generate code completion using Yuuki."""
+    
+    if not MODEL_LOADED:
+        if not load_model():
+            return "Error: Model failed to load. Please try refreshing the page."
     
     if not prompt or not prompt.strip():
         return "Please enter a code prompt."
     
     try:
-        from huggingface_hub import InferenceClient
-        
-        client = InferenceClient()
-        response = client.text_generation(
-            prompt,
-            model=MODEL_ID,
-            max_new_tokens=max_new_tokens,
-            temperature=temperature,
-            top_p=top_p,
-            do_sample=True
+        inputs = tokenizer(
+            prompt, 
+            return_tensors="pt",
+            truncation=True,
+            max_length=512
         )
         
-        return response
+        with torch.no_grad():
+            outputs = model.generate(
+                **inputs,
+                max_new_tokens=max_new_tokens,
+                temperature=temperature,
+                top_p=top_p,
+                top_k=top_k,
+                repetition_penalty=repetition_penalty,
+                do_sample=True,
+                pad_token_id=tokenizer.pad_token_id,
+                eos_token_id=tokenizer.eos_token_id,
+                num_return_sequences=1
+            )
         
+        generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        return generated_text
+    
     except Exception as e:
-        return f"Generation error: {str(e)}\n\nTry model directly: https://huggingface.co/OpceanAI/Yuuki-best"
+        return f"Generation error: {str(e)}"
+
 
 # ============================================================================
 # Examples
@@ -633,9 +682,15 @@ with gr.Blocks(
             # Examples
             gr.HTML('<div id="examples-label">Try these</div>')
             with gr.Row(elem_id="examples-grid"):
-                for ex in EXAMPLES:
-                    btn = gr.Button(ex[0], elem_classes=["example-btn"], size="sm")
-                    btn.click(lambda x=ex[0]: x, outputs=prompt_input)
+                ex_btn_1 = gr.Button("module Main where", elem_classes=["example-btn"], size="sm")
+                ex_btn_2 = gr.Button("open import Data.Nat", elem_classes=["example-btn"], size="sm")
+                ex_btn_3 = gr.Button("int main() {", elem_classes=["example-btn"], size="sm")
+                ex_btn_4 = gr.Button("def hello():", elem_classes=["example-btn"], size="sm")
+            
+            ex_btn_1.click(lambda: "module Main where", outputs=prompt_input)
+            ex_btn_2.click(lambda: "open import Data.Nat", outputs=prompt_input)
+            ex_btn_3.click(lambda: "int main() {", outputs=prompt_input)
+            ex_btn_4.click(lambda: "def hello():", outputs=prompt_input)
         
         # ===== SETTINGS TAB =====
         with gr.Tab("Settings", id="settings"):
@@ -729,14 +784,4 @@ with gr.Blocks(
                 gr.HTML("""
                     <div class="score-grid">
                         <span class="score-badge good">Agda: 55/100</span>
-                        <span class="score-badge medium">C: 20/100</span>
-                        <span class="score-badge medium">Assembly: 15/100</span>
-                        <span class="score-badge weak">Python: 8/100</span>
-                    </div>
-                    <p style="color: #666; font-size: 0.8rem; margin-top: 16px; line-height: 1.5;">
-                        Python scores low due to alphabetical dataset ordering. 
-                        Average quality: 24.6/100 (+146% from checkpoint 1400).
-                    </p>
-                """)
-            
-        
+                        <span class="score-badge medium">C: 20/100</sp

requirements.txt

@@ -1,4 +1,5 @@
-gradio==5.9.1
-transformers==4.46.0
-torch>=2.5.0
-accelerate
+gradio==4.44.1
+transformers==4.45.0
+torch==2.5.0
+accelerate==0.34.0
+huggingface-hub>=0.20.0