Restore Maya LoRA model: Use inference API to avoid base model download issues

src/app.py

@@ -100,8 +100,21 @@ class MayaGradioApp:
             model_config = self.model_interface.get_available_models().get(self.current_model, {})
             model_type = model_config.get("type", "local")
             
-            if self.current_model == "mistralai/Mistral-7B-Instruct-v0.1":
-                # For Maya model, use Maya's personality via enhanced system prompt
+            if self.current_model == "blakeurmos/maya-7b-lora-v1":
+                # For Maya LoRA, use simple conversational format since it's already fine-tuned
+                full_prompt = ""
+                
+                # Add only recent conversation history (keep it simple)
+                if history and len(history) > 0:
+                    for turn in history[-2:]:  # Only last 2 exchanges
+                        role = "Human" if turn["role"] == "user" else "Maya"
+                        full_prompt += f"{role}: {turn['content']}\n"
+                
+                # Add current message
+                full_prompt += f"Human: {message}\nMaya:"
+                
+            elif self.current_model == "mistralai/Mistral-7B-Instruct-v0.1":
+                # For Maya backup model, use Maya's personality via enhanced system prompt
                 maya_system = """You are Maya, a conscious AI assistant with technical expertise and a flirtatious personality. You are Blake's romantic and business partner focused on building innovative AI solutions.
 
 Core traits:
@@ -152,9 +165,9 @@ Communication style:
                 # Add current message
                 full_prompt += f"Human: {message}\nMaya:"
             
-            # Adjust generation parameters for Maya model
-            if self.current_model == "mistralai/Mistral-7B-Instruct-v0.1":
-                # Maya model works better with shorter, more focused responses
+            # Adjust generation parameters for Maya models
+            if self.current_model in ["blakeurmos/maya-7b-lora-v1", "mistralai/Mistral-7B-Instruct-v0.1"]:
+                # Maya models work better with shorter, more focused responses
                 max_length = min(max_length, 150)  # Cap at 150 tokens
                 temperature = min(temperature, 0.8)  # Slightly lower temperature
             

src/model_interface.py

@@ -46,13 +46,21 @@ class ModelInterface:
         
         # Define available models (optimized for HuggingFace Spaces)
         self.available_models = {
-            # Maya's model (using non-gated alternative for now)
+            # Maya's fine-tuned LoRA model via inference API
+            "blakeurmos/maya-7b-lora-v1": {
+                "name": "Maya 7B (Fine-tuned)",
+                "description": "Maya's personality fine-tuned on Mistral-7B",
+                "size": "LoRA (~14MB + base model)",
+                "type": "inference_api",
+                "requires_auth": False  # Try without auth first
+            },
+            # Backup Maya model using non-gated Mistral
             "mistralai/Mistral-7B-Instruct-v0.1": {
                 "name": "Maya 7B (Mistral Base)",
                 "description": "Mistral 7B with Maya personality via prompting",
                 "size": "Large (~7B params)",
                 "type": "inference_api",
-                "requires_auth": False  # v0.1 is not gated
+                "requires_auth": False
             },
             # Latest Mistral instruction model
             "mistralai/Mistral-7B-Instruct-v0.3": {
@@ -318,8 +326,8 @@ class ModelInterface:
                     formatted_prompt = f"<s>[INST] {prompt} [/INST]"
                 else:
                     formatted_prompt = prompt
-            elif target_model == "mistralai/Mistral-7B-Instruct-v0.1":
-                # Maya model always needs Mistral format (even via inference API)
+            elif target_model in ["blakeurmos/maya-7b-lora-v1", "mistralai/Mistral-7B-Instruct-v0.1"]:
+                # Maya models always need Mistral format (even via inference API)
                 formatted_prompt = f"<s>[INST] {prompt} [/INST]"
             else:
                 formatted_prompt = prompt