update

deciders/utils.py

@@ -39,6 +39,22 @@ def get_completion(prompt: str, api_type: str = "azure", engine: str = "gpt-35-t
                     # request_timeout = 1
                 )
         return response.choices[0].text
+    elif api_type == "openai":
+        messages = [
+            {
+                "role": "user",
+                "content": prompt
+            }
+        ]
+        response = openai.ChatCompletion.create(
+            model=engine,
+            messages=messages,
+            max_tokens=max_tokens,
+            stop=stop_strs,
+            temperature=temperature,
+            # request_timeout = 1
+        )
+        return response.choices[0]["message"]["content"]
 
 # @retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
 def get_chat(prompt: str, api_type: str = "azure", model: str = "gpt-35-turbo", engine: str = "gpt-35-turbo", temperature: float = 0.0, max_tokens: int = 256, stop_strs: Optional[List[str]] = None, is_batched: bool = False) -> str:
@@ -60,4 +76,14 @@ def get_chat(prompt: str, api_type: str = "azure", model: str = "gpt-35-turbo",
             # request_timeout = 1
         )
         return response.choices[0]["message"]["content"]
+    elif api_type == "openai":
+        response = openai.ChatCompletion.create(
+            model=model,
+            messages=messages,
+            max_tokens=max_tokens,
+            stop=stop_strs,
+            temperature=temperature,
+            # request_timeout = 1
+        )
+        return response.choices[0]["message"]["content"]
 

distillers/self_reflection.py

@@ -45,7 +45,7 @@ class RefletionGenerator():
             for i, m in enumerate(memory):
                 query += f'Trial #{i}: {m}\n'
 
-        query += '\n\nNew plan:'
+        query += '\n\nPlease give your new plan.'
         return query
 
     def generate(self, traj, memory, max_len_mem=5):

main_reflexion.py

@@ -99,7 +99,7 @@ def _run(translator, environment, decider, max_episode_len, logfile, args, trail
         error_flag = True
         retry_num = 1
         for error_i in range(retry_num):
-            # try:
+            try:
                 action, prompt, response, tokens, cost = decider.act(
                     state_description,
                     action_description,
@@ -127,15 +127,15 @@ def _run(translator, environment, decider, max_episode_len, logfile, args, trail
                 current_total_cost += cost
                 error_flag = False
                 break
-            # except Exception as e:
-            #     print(e)
-            #     if error_i < retry_num-1:
-            #         if "Cliff" in args.env_name or "Frozen" in args.env_name:
-            #             decider.env_history.remove_invalid_state()
-            #         decider.env_history.remove_invalid_state()
-            #     if logger:
-            #         logger.debug(f"Error: {e}, Retry! ({error_i+1}/{retry_num})")
-            #     continue
+            except Exception as e:
+                print(e)
+                if error_i < retry_num-1:
+                    if "Cliff" in args.env_name or "Frozen" in args.env_name:
+                        decider.env_history.remove_invalid_state()
+                    decider.env_history.remove_invalid_state()
+                if logger:
+                    logger.debug(f"Error: {e}, Retry! ({error_i+1}/{retry_num})")
+                continue
         if error_flag:
             if "Continuous" in args.env_name:
                 action = [decider.default_action]