Fixed Model references after refactor to r1 1776

app.py

@@ -350,13 +350,10 @@ def get_current_time_in_timezone(timezone: str) -> str:
 
 final_answer = FinalAnswerTool()
 
-# Replace current model with Perplexity AI R1-1776 (128K context window)
+# Create Perplexity R1 model implementation directly without referencing an undefined variable
 
-# Import additional necessary modules
-from huggingface_hub import InferenceClient
-
-# Keep the original model definition but don't use it
-original_model = model
+# Import necessary modules (already imported above)
+# from huggingface_hub import InferenceClient
 
 # Create a new model implementation that uses the larger context window model through InferenceClient
 class PerplexityR1Model:
@@ -397,12 +394,12 @@ class PerplexityR1Model:
             return completion.choices[0].message.content
         except Exception as e:
             print(f"Error calling Perplexity R1-1776: {str(e)}")
-            # If we get an error with the large context model, try our aggressive context trimming
+            # For context length errors, try simple truncation
             if "context length" in str(e).lower() or "token limit" in str(e).lower():
-                print("Context length error with R1-1776 - trimming context and retrying")
-                # Use our existing context management function
-                trimmed_prompt = manage_context(prompt, max_allowed_tokens=90000)  # 90K to be safe
-                messages = [{"role": "user", "content": trimmed_prompt}]
+                print("Context length error with R1-1776 - truncating prompt and retrying")
+                # Simple truncation - take the last ~80K characters which should be under the limit
+                truncated_prompt = prompt[-80000:] if len(prompt) > 80000 else prompt
+                messages = [{"role": "user", "content": truncated_prompt}]
                 
                 try:
                     completion = self.client.chat.completions.create(
@@ -414,19 +411,22 @@ class PerplexityR1Model:
                     return completion.choices[0].message.content
                 except Exception as retry_error:
                     print(f"Error on retry: {str(retry_error)}")
-                    # Fall back to error message
-                    return f"ERROR: Model call failed even with reduced context. Please try a shorter query. Error: {str(retry_error)}"
+                    return f"ERROR: Model call failed even with reduced context. Please try a shorter query."
             else:
-                # For non-context errors, return error message
+                # For non-context errors
                 return f"ERROR: {str(e)}"
 
-# Replace the model with our new implementation
+# Initialize our model with Perplexity R1-1776
 model = PerplexityR1Model(temperature=0.5, max_tokens=1500)
 
-# No need for complex context management or fallbacks now with the large context window
-# But keep the functions in place in case they're needed as fallbacks
+# Import tool from Hub - do this before using the tool in the agent
+image_generation_tool = load_tool("agents-course/text-to-image", trust_remote_code=True)
+
+# Load prompt templates before using them in the agent
+with open("prompts.yaml", 'r') as stream:
+    prompt_templates = yaml.safe_load(stream)
 
-# Update the agent with the new model and more steps
+# Initialize the agent with all required components already defined
 agent = CodeAgent(
     model=model,
     tools=[
@@ -439,8 +439,8 @@ agent = CodeAgent(
         Check_Dataset_Validity,
         visit_webpage_tool,
     ],
-    max_steps=12,  # Increase back to 12 since we have a large context window
-    verbosity_level=1,  # Increase to 1 since we have room
+    max_steps=12,
+    verbosity_level=1,
     grammar=None,
     planning_interval=2,
     name="Research Assistant",
@@ -452,12 +452,6 @@ agent = CodeAgent(
 # Add informative message about the model
 print("Using Perplexity R1-1776 model with 128K token context window")
 
-# Import tool from Hub
-image_generation_tool = load_tool("agents-course/text-to-image", trust_remote_code=True)
-
-with open("prompts.yaml", 'r') as stream:
-    prompt_templates = yaml.safe_load(stream)
-
 # 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")