Fix config override bug by recreating LMs after load_state

nanocode.py

@@ -170,8 +170,8 @@ def run_bash(cmd: str) -> str:
 class RLMReasoningCallback(BaseCallback):
     def on_module_end(self, call_id, outputs, exception):
         if outputs and hasattr(outputs, "reasoning") and hasattr(outputs, "code"):
-            print(f"{DIM}⏺ [REASONING STEP] {outputs.reasoning}\n{RESET}")
-            print(f"{DIM}⏺ [CODE] {outputs.code}\n{RESET}")
+            print(f"{DIM}⏺ [REASONING STEP]\n{outputs.reasoning}\n{RESET}")
+            print(f"{DIM}⏺ [CODE]\n```\n{outputs.code}\n```\n{RESET}")
 
 
 # -- Program ---
@@ -193,7 +193,7 @@ class RLMCodingConfig(PrecompiledConfig):
     api_base: str = "https://openrouter.ai/api/v1"
     max_tokens: int = 50000
     max_output_chars: int = 100000
-    verbose: bool = False
+    verbose: bool = True
     track_usage: bool = True
 
 
@@ -261,22 +261,22 @@ class RLMCodingProgram(PrecompiledProgram):
 
             def wrapped_q(prompt):  # wrap query
                 print(
-                    f"{DIM}⏺ [LLM QUERY]: {prompt[:100]}...{RESET}\n"
+                    f"{DIM}⏺ [LLM QUERY]:\n{prompt[:100]}...{RESET}\n"
                     if len(prompt) > 100
-                    else f"{DIM}⏺ [LLM QUERY]: {prompt}{RESET}\n"
+                    else f"{DIM}⏺ [LLM QUERY]:\n{prompt}{RESET}\n"
                 )
                 res = orig_q(prompt)
                 print(
-                    f"{DIM}⏺ [LLM QUERY RESULT]: {str(res)[:200]}...{RESET}\n"
+                    f"{DIM}⏺ [LLM QUERY RESULT]:\n{str(res)[:200]}...{RESET}\n"
                     if len(str(res)) > 200
-                    else f"{DIM}⏺ [LLM QUERY RESULT]: {res}{RESET}\n"
+                    else f"{DIM}⏺ [LLM QUERY RESULT]:\n{res}{RESET}\n"
                 )
                 return res
 
             def wrapped_b(prompts):  # wrap batched query
-                print(f"{DIM}⏺ [LLM QUERY BATCHED]: {len(prompts)} prompts{RESET}\n")
+                print(f"{DIM}⏺ [LLM QUERY BATCHED]:\n{len(prompts)} prompts{RESET}\n")
                 res = orig_b(prompts)
-                print(f"{DIM}⏺ [LLM QUERY BATCHED]: {len(res)} results{RESET}\n")
+                print(f"{DIM}⏺ [LLM QUERY BATCHED]:\n{len(res)} results{RESET}\n")
                 return res
 
             tools["llm_query"] = wrapped_q
@@ -318,7 +318,7 @@ class RLMCodingProgram(PrecompiledProgram):
             tools=self.tools,
             max_output_chars=self.config.max_output_chars,
             max_iterations=self.config.max_iters,
-            verbose=self.config.verbose,
+            verbose=False,  # We add our own verbose logging
         )
         new_instance.set_lm(self.lm)
         self.agent = new_instance
@@ -358,6 +358,7 @@ class RLMCodingProgram(PrecompiledProgram):
 
 if __name__ == "__main__":
     agent = RLMCodingProgram(RLMCodingConfig())
+    #agent(task="explicity call llm_query(who is the ceo of apple?) to get the answer to 'who is the ceo of apple?'")
     branches = ["main", "dev", "prod"]
     for branch in branches:
         agent.push_to_hub(
@@ -366,3 +367,4 @@ if __name__ == "__main__":
             branch=branch,
         )
 
+