Updated api llm endpoint to deepseek

app.py

@@ -350,67 +350,100 @@ def get_current_time_in_timezone(timezone: str) -> str:
 
 final_answer = FinalAnswerTool()
 
-# Remove the huggingface_api_key parameter - it's not supported
-model = HfApiModel(
+# Define the primary DeepSeek model using InferenceClient
+def primary_deepseek_model(prompt):
+    """Use the DeepSeek-R1-Distill-Qwen-32B model as the primary endpoint."""
+    try:
+        print("Using primary model: DeepSeek-R1-Distill-Qwen-32B via HF Inference API")
+        # Get API key
+        api_key = os.getenv("HF_API_KEY") or os.getenv("HUGGINGFACE_API_KEY")
+        if not api_key:
+            raise ValueError("No Hugging Face API key found in environment variables")
+            
+        client = InferenceClient(
+            provider="hf-inference",
+            api_key=api_key
+        )
+        
+        messages = [{"role": "user", "content": prompt}]
+        completion = client.chat.completions.create(
+            model="deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", 
+            messages=messages, 
+            max_tokens=2096,
+            temperature=0.5
+        )
+        print("Primary model call successful")
+        return completion.choices[0].message.content
+    except Exception as e:
+        print(f"Primary DeepSeek model failed: {str(e)}")
+        raise e
+
+# Keep the original endpoint as a backup
+backup_model = HfApiModel(
     max_tokens=2096,
     temperature=0.5,
-    model_id='https://pflgm2locj2t89co.us-east-1.aws.endpoints.huggingface.cloud',  # Using the backup endpoint
+    model_id='https://pflgm2locj2t89co.us-east-1.aws.endpoints.huggingface.cloud',
     custom_role_conversions=None
 )
 
-# Add fallback logic that only activates if the primary model fails
+# Updated fallback mechanism that tries primary first, then original model, then additional fallbacks
 def try_model_call_with_fallbacks(prompt):
-    """Try to use the primary model first, fall back to alternatives if it fails."""
-    # First attempt with primary model
+    """Try to use the primary model first, fall back to original model, then try other fallbacks if both fail."""
+    # First attempt with primary DeepSeek model
     try:
-        return model(prompt)
+        return primary_deepseek_model(prompt)
     except Exception as primary_error:
         print(f"Primary model call failed: {str(primary_error)}")
-        print("Trying fallback models...")
+        print("Trying backup model...")
         
-        # List of fallback models
-        fallbacks = [
-            {
-                "provider": "sambanova",
-                "model_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
-                "display_name": "Qwen 2.5 Coder 32B"
-            },
-            {
-                "provider": "hf-inference",
-                "model_name": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
-                "display_name": "DeepSeek R1 Distill Qwen 32B"
-            }
-        ]
-        
-        # Get API key
-        api_key = os.getenv("HF_API_KEY") or os.getenv("HUGGINGFACE_API_KEY")
-        if not api_key:
-            raise ValueError("No Hugging Face API key found in environment variables")
-        
-        # Try each fallback model in sequence
-        for fallback in fallbacks:
-            try:
-                print(f"Trying fallback model: {fallback['display_name']}")
-                client = InferenceClient(provider=fallback["provider"], api_key=api_key)
-                messages = [{"role": "user", "content": prompt}]
-                completion = client.chat.completions.create(
-                    model=fallback["model_name"],
-                    messages=messages,
-                    max_tokens=2096,
-                    temperature=0.5
-                )
-                print(f"Successfully used fallback model: {fallback['display_name']}")
-                return completion.choices[0].message.content
-            except Exception as e:
-                print(f"Fallback model {fallback['display_name']} failed: {str(e)}")
-                continue
-        
-        # If all fallbacks fail, re-raise the original error
-        raise primary_error
+        # Second attempt with original backup model
+        try:
+            print("Using backup model: HfApiModel endpoint")
+            result = backup_model(prompt)
+            print("Backup model call successful")
+            return result
+        except Exception as backup_error:
+            print(f"Backup model call failed: {str(backup_error)}")
+            print("Trying additional fallback models...")
+            
+            # List of additional fallback models
+            fallbacks = [
+                {
+                    "provider": "sambanova",
+                    "model_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
+                    "display_name": "Qwen 2.5 Coder 32B"
+                }
+                # DeepSeek is now the primary model, so we removed it from fallbacks
+            ]
+            
+            # Get API key
+            api_key = os.getenv("HF_API_KEY") or os.getenv("HUGGINGFACE_API_KEY")
+            if not api_key:
+                raise ValueError("No Hugging Face API key found in environment variables")
+            
+            # Try each fallback model in sequence
+            for fallback in fallbacks:
+                try:
+                    print(f"Trying fallback model: {fallback['display_name']}")
+                    client = InferenceClient(provider=fallback["provider"], api_key=api_key)
+                    messages = [{"role": "user", "content": prompt}]
+                    completion = client.chat.completions.create(
+                        model=fallback["model_name"],
+                        messages=messages,
+                        max_tokens=2096,
+                        temperature=0.5
+                    )
+                    print(f"Successfully used fallback model: {fallback['display_name']}")
+                    return completion.choices[0].message.content
+                except Exception as e:
+                    print(f"Fallback model {fallback['display_name']} failed: {str(e)}")
+                    continue
+            
+            # If all fallbacks fail, re-raise the original error
+            raise primary_error
 
-# Monkey patch the model's __call__ method to use our fallback logic
-original_call = model.__call__
-model.__call__ = try_model_call_with_fallbacks
+# Create a model wrapper that uses our fallback logic
+model = lambda prompt: try_model_call_with_fallbacks(prompt)
 
 # Import tool from Hub
 image_generation_tool = load_tool("agents-course/text-to-image", trust_remote_code=True)
@@ -447,6 +480,7 @@ agent = CodeAgent(
 # Add informative message about which search tool is being used
 print(f"Agent initialized with {search_tool_name} as primary search tool")
 print(f"Available tools: final_answer, Sonar_Web_Search_Tool, {search_tool_name}, get_current_time_in_timezone, image_generation_tool, Dataset_Creator_Tool, Check_Dataset_Validity, visit_webpage_tool")
+print(f"Using DeepSeek-R1-Distill-Qwen-32B as primary model, with HfApiModel as backup")
 
 # Note about working within token limits - add this comment
 # When using with queries that might exceed token limits, consider: