Update app.py

app.py

@@ -314,11 +314,15 @@ COHERE_MODELS = {
 
 # TOGETHER MODELS in the free tier
 TOGETHER_MODELS = {
+    "meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo": 131072,
     "deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free": 8192,
     "meta-llama/Llama-Vision-Free": 8192,
     "meta-llama/Llama-3.3-70B-Instruct-Turbo-Free": 8192,
 }
 
+# Add to the vision models list
+VISION_MODELS["Together"] = ["meta-llama/Llama-Vision-Free"]
+
 # OVH MODELS - OVH AI Endpoints (free beta)
 OVH_MODELS = {
     "ovh/codestral-mamba-7b-v0.1": 131072,
@@ -501,7 +505,7 @@ def prepare_message_with_media(text, images=None, documents=None):
             return text
     
     # If we have images, create a multimodal content array
-    content = [{"type": "text", "text": text}]
+    content = [{"type": "text", "text": text or "Analyze this image:"}]
     
     # Add images if any
     if images:
@@ -911,8 +915,14 @@ def extract_ai_response(result, provider):
                 return "No response content from Cohere"
                 
         elif provider == "Together":
+            # Handle response from Together's native client
             if hasattr(result, "choices") and len(result.choices) > 0:
-                return result.choices[0].message.content
+                if hasattr(result.choices[0], "message") and hasattr(result.choices[0].message, "content"):
+                    return result.choices[0].message.content
+                elif hasattr(result.choices[0], "delta") and hasattr(result.choices[0].delta, "content"):
+                    return result.choices[0].delta.content
+            # Fallback
+            return str(result)
                 
         elif provider == "OVH":
             if isinstance(result, dict) and "choices" in result and len(result["choices"]) > 0:
@@ -933,27 +943,32 @@ def extract_ai_response(result, provider):
         return f"Error: {str(e)}"
 
 def call_together_api(payload, api_key_override=None):
-    """Make a call to Together API with error handling"""
+    """Make a call to Together API with error handling using their native client"""
     try:
-        if not HAS_OPENAI:
-            raise ImportError("OpenAI package not installed (required for Together API)")
-            
+        # Import Together's native client
+        # Note: This might need to be installed with: pip install together
+        try:
+            from together import Together
+        except ImportError:
+            raise ImportError("The Together Python package is not installed. Please install it with: pip install together")
+        
         api_key = api_key_override if api_key_override else TOGETHER_API_KEY
         if not api_key:
             raise ValueError("Together API key is required")
         
-        # Create client with Together base URL
-        client = openai.OpenAI(
-            api_key=api_key,
-            base_url="https://api.together.xyz/v1"
-        )
+        # Create the Together client
+        client = Together(api_key=api_key)
         
         # Extract parameters from payload
         requested_model = payload.get("model", "")
+        messages = payload.get("messages", [])
+        temperature = payload.get("temperature", 0.7)
+        max_tokens = payload.get("max_tokens", 1000)
+        stream = payload.get("stream", False)
         
-        # Use a safe model that's known to work in the free tier
-        # these models are available without dedicated endpoints
+        # Use one of the free, serverless models
         free_models = [
+            "meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
             "deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free",
             "meta-llama/Llama-Vision-Free",
             "meta-llama/Llama-3.3-70B-Instruct-Turbo-Free"
@@ -965,31 +980,41 @@ def call_together_api(payload, api_key_override=None):
         # Try to match a requested model with a free model if possible
         if requested_model:
             for free_model in free_models:
-                if requested_model.lower() in free_model.lower():
+                if requested_model.lower() in free_model.lower() or free_model.lower() in requested_model.lower():
                     model = free_model
                     break
         
-        # Create payload with clean messages
-        messages = []
-        for msg in payload.get("messages", []):
-            # Ensure we only include role and content
-            clean_msg = {
-                "role": msg["role"],
-                "content": msg["content"]
-            }
-            messages.append(clean_msg)
+        # Process messages for possible image content
+        processed_messages = []
+        for msg in messages:
+            role = msg["role"]
+            content = msg["content"]
             
-        # Create payload
-        together_payload = {
-            "model": model,
-            "messages": messages,
-            "temperature": payload.get("temperature", 0.7),
-            "max_tokens": payload.get("max_tokens", 1000),
-            "stream": payload.get("stream", False)
-        }
+            # Handle multimodal content for vision models
+            if isinstance(content, list) and "vision" in model.lower():
+                # Format according to Together's expected multimodal format
+                parts = []
+                for item in content:
+                    if item["type"] == "text":
+                        parts.append({"type": "text", "text": item["text"]})
+                    elif item["type"] == "image_url":
+                        parts.append({
+                            "type": "image_url",
+                            "image_url": item["image_url"]
+                        })
+                processed_messages.append({"role": role, "content": parts})
+            else:
+                # Regular text messages
+                processed_messages.append({"role": role, "content": content})
         
-        # Create completion
-        response = client.chat.completions.create(**together_payload)
+        # Create completion with Together's client
+        response = client.chat.completions.create(
+            model=model,
+            messages=processed_messages,
+            temperature=temperature,
+            max_tokens=max_tokens,
+            stream=stream
+        )
         
         return response
     except Exception as e: