update

deciders/act.py

@@ -6,7 +6,7 @@ from loguru import logger
 from .parser import PARSERS
 from langchain.output_parsers import PydanticOutputParser
 from langchain.output_parsers import OutputFixingParser
-from langchain.chat_models import AzureChatOpenAI
+from langchain.chat_models import AzureChatOpenAI, ChatOpenAI
 from memory.env_history import EnvironmentHistory
 import tiktoken
 import json
@@ -88,15 +88,18 @@ class NaiveAct(gpt):
         else:
             num_action = 1
 
-        autofixing_chat = AzureChatOpenAI(
-            openai_api_type=openai.api_type,
-            openai_api_version=openai.api_version,
-            openai_api_base=openai.api_base,
-            openai_api_key=openai.api_key,
-            deployment_name=self.args.gpt_version,
-            temperature=self.temperature,
-            max_tokens=self.max_tokens
-        )
+        if self.args.api_type == "azure":
+            autofixing_chat = AzureChatOpenAI(
+                openai_api_type=openai.api_type,
+                openai_api_version=openai.api_version,
+                openai_api_base=openai.api_base,
+                openai_api_key=openai.api_key,
+                deployment_name=self.args.gpt_version,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens
+            )
+        elif self.args.api_type == "openai":
+            autofixing_chat = ChatOpenAI(temperature=self.temperature, openai_api_key=openai.api_key)
 
         parser = PydanticOutputParser(pydantic_object=PARSERS[num_action])
         autofixing_parser = OutputFixingParser.from_llm(

deciders/cot.py

@@ -1,6 +1,6 @@
 import openai
 from .misc import history_to_str
-from langchain.chat_models import AzureChatOpenAI
+from langchain.chat_models import AzureChatOpenAI, ChatOpenAI
 from langchain.prompts.chat import (
     PromptTemplate,
     ChatPromptTemplate,
@@ -31,15 +31,20 @@ class ChainOfThought(NaiveAct):
     ):
         self.action_description = action_description
         self._add_history_before_action(game_description, goal_description, state_description)
-        chat = AzureChatOpenAI(
-            openai_api_type=openai.api_type,
-            openai_api_version=openai.api_version,
-            openai_api_base=openai.api_base,
-            openai_api_key=openai.api_key,
-            deployment_name=self.args.gpt_version,
-            temperature=self.temperature,
-            max_tokens=self.max_tokens
-        )
+        
+        if self.args.api_type == "azure":
+            chat = AzureChatOpenAI(
+                openai_api_type=openai.api_type,
+                openai_api_version=openai.api_version,
+                openai_api_base=openai.api_base,
+                openai_api_key=openai.api_key,
+                deployment_name=self.args.gpt_version,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens
+            )
+        elif self.args.api_type == "openai":
+            chat = ChatOpenAI(temperature=self.temperature, openai_api_key=openai.api_key)
+    
 
         suffix_flag = False
         reply_format_description = \

deciders/exe.py

@@ -1,6 +1,6 @@
 import openai
 from .misc import history_to_str
-from langchain.chat_models import AzureChatOpenAI
+from langchain.chat_models import AzureChatOpenAI, ChatOpenAI
 from langchain.prompts.chat import (
     PromptTemplate,
     ChatPromptTemplate,
@@ -96,15 +96,19 @@ class EXE(NaiveAct):
         self.game_description = game_description 
         self.goal_description = goal_description
         self.env_history.add("observation", state_description)
-        chat = AzureChatOpenAI(
-            openai_api_type=openai.api_type,
-            openai_api_version=openai.api_version,
-            openai_api_base=openai.api_base,
-            openai_api_key=openai.api_key,
-            deployment_name=self.args.gpt_version,
-            temperature=self.temperature,
-            max_tokens=self.max_tokens,
-        )
+
+        if self.args.api_type == "azure":
+            chat = AzureChatOpenAI(
+                openai_api_type=openai.api_type,
+                openai_api_version=openai.api_version,
+                openai_api_base=openai.api_base,
+                openai_api_key=openai.api_key,
+                deployment_name=self.args.gpt_version,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens
+            )
+        elif self.args.api_type == "openai":
+            chat = ChatOpenAI(temperature=self.temperature, openai_api_key=openai.api_key)
         # print(self.logger)
         reply_format_description = \
             "Your response should choose an optimal action from valid action list, and terminated with following format: "        

deciders/reflexion.py

@@ -1,6 +1,6 @@
 import openai
 from .misc import history_to_str
-from langchain.chat_models import AzureChatOpenAI
+from langchain.chat_models import AzureChatOpenAI, ChatOpenAI
 from langchain.prompts.chat import (
     PromptTemplate,
     ChatPromptTemplate,
@@ -53,15 +53,19 @@ class Reflexion(NaiveAct):
         self.game_description = game_description 
         self.goal_description = goal_description
         self.env_history.add("observation", state_description)
-        chat = AzureChatOpenAI(
-            openai_api_type=openai.api_type,
-            openai_api_version=openai.api_version,
-            openai_api_base=openai.api_base,
-            openai_api_key=openai.api_key,
-            deployment_name=self.args.gpt_version,
-            temperature=self.temperature,
-            max_tokens=self.max_tokens,
-        )
+
+        if self.args.api_type == "azure":
+            chat = AzureChatOpenAI(
+                openai_api_type=openai.api_type,
+                openai_api_version=openai.api_version,
+                openai_api_base=openai.api_base,
+                openai_api_key=openai.api_key,
+                deployment_name=self.args.gpt_version,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens
+            )
+        elif self.args.api_type == "openai":
+            chat = ChatOpenAI(temperature=self.temperature, openai_api_key=openai.api_key)
         suffix_flag = False
         reply_format_description = \
             "Your response should choose an optimal action from a valid action list and terminate with the following format: "

deciders/self_consistency.py

@@ -1,6 +1,6 @@
 import openai
 from .misc import history_to_str
-from langchain.chat_models import AzureChatOpenAI
+from langchain.chat_models import AzureChatOpenAI, ChatOpenAI
 from langchain.prompts.chat import (
     PromptTemplate,
     ChatPromptTemplate,
@@ -34,15 +34,19 @@ class SelfConsistency(NaiveAct):
         # print(self.temperature)
         self.action_description = action_description
         self._add_history_before_action(game_description, goal_description, state_description)
-        chat = AzureChatOpenAI(
-            openai_api_type=openai.api_type,
-            openai_api_version=openai.api_version,
-            openai_api_base=openai.api_base,
-            openai_api_key=openai.api_key,
-            deployment_name=self.args.gpt_version,
-            temperature=self.temperature,
-            max_tokens=self.max_tokens
-        )
+
+        if self.args.api_type == "azure":
+            chat = AzureChatOpenAI(
+                openai_api_type=openai.api_type,
+                openai_api_version=openai.api_version,
+                openai_api_base=openai.api_base,
+                openai_api_key=openai.api_key,
+                deployment_name=self.args.gpt_version,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens
+            )
+        elif self.args.api_type == "openai":
+            chat = ChatOpenAI(temperature=self.temperature, openai_api_key=openai.api_key)
 
         suffix_flag = False
         reply_format_description = \

deciders/selfask.py

@@ -1,6 +1,6 @@
 import openai
 from .misc import history_to_str
-from langchain.chat_models import AzureChatOpenAI
+from langchain.chat_models import AzureChatOpenAI, ChatOpenAI
 from langchain.prompts.chat import (
     PromptTemplate,
     ChatPromptTemplate,
@@ -31,15 +31,19 @@ class SelfAskAct(NaiveAct):
     ):
         self.action_description = action_description
         self._add_history_before_action(game_description, goal_description, state_description)
-        chat = AzureChatOpenAI(
-            openai_api_type=openai.api_type,
-            openai_api_version=openai.api_version,
-            openai_api_base=openai.api_base,
-            openai_api_key=openai.api_key,
-            deployment_name=self.args.gpt_version,
-            temperature=self.temperature,
-            max_tokens=self.max_tokens
-        )
+
+        if self.args.api_type == "azure":
+            chat = AzureChatOpenAI(
+                openai_api_type=openai.api_type,
+                openai_api_version=openai.api_version,
+                openai_api_base=openai.api_base,
+                openai_api_key=openai.api_key,
+                deployment_name=self.args.gpt_version,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens
+            )
+        elif self.args.api_type == "openai":
+            chat = ChatOpenAI(temperature=self.temperature, openai_api_key=openai.api_key)
 
         suffix_flag = False
         reply_format_description = \

deciders/spp.py

@@ -1,6 +1,6 @@
 import openai
 from .misc import history_to_str
-from langchain.chat_models import AzureChatOpenAI
+from langchain.chat_models import AzureChatOpenAI, ChatOpenAI
 from langchain.prompts.chat import (
     PromptTemplate,
     ChatPromptTemplate,
@@ -30,15 +30,19 @@ class SPP(NaiveAct):
     ):
         self.action_description = action_description
         self._add_history_before_action(game_description, goal_description, state_description)
-        chat = AzureChatOpenAI(
-            openai_api_type=openai.api_type,
-            openai_api_version=openai.api_version,
-            openai_api_base=openai.api_base,
-            openai_api_key=openai.api_key,
-            deployment_name=self.args.gpt_version,
-            temperature=self.temperature,
-            max_tokens=self.max_tokens
-        )
+
+        if self.args.api_type == "azure":
+            chat = AzureChatOpenAI(
+                openai_api_type=openai.api_type,
+                openai_api_version=openai.api_version,
+                openai_api_base=openai.api_base,
+                openai_api_key=openai.api_key,
+                deployment_name=self.args.gpt_version,
+                temperature=self.temperature,
+                max_tokens=self.max_tokens
+            )
+        elif self.args.api_type == "openai":
+            chat = ChatOpenAI(temperature=self.temperature, openai_api_key=openai.api_key)
 
         self.fewshot_example = self.irr_few_shot_examples if not self.fewshot_example else self.fewshot_example
         self.irr_few_shot_examples = self.irr_few_shot_examples if not self.fewshot_example else self.fewshot_example

deciders/utils.py

@@ -1,6 +1,6 @@
 import os
 import sys
-import openai
+import openai # 0.27.8
 from tenacity import (
     retry,
     stop_after_attempt, # type: ignore
@@ -24,9 +24,10 @@ import timeout_decorator
 def run_chain(chain, *args, **kwargs):
     return chain.run(*args, **kwargs)
 
-@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
-def get_completion(prompt: str, engine: str = "gpt-35-turbo", temperature: float = 0.0, max_tokens: int = 256, stop_strs: Optional[List[str]] = None) -> str:
+# @retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
+def get_completion(prompt: str, api_type: str = "azure", engine: str = "gpt-35-turbo", temperature: float = 0.0, max_tokens: int = 256, stop_strs: Optional[List[str]] = None) -> str:
     response = openai.Completion.create(
+                model=engine,
                 engine=engine,
                 prompt=prompt,
                 temperature=temperature,
@@ -39,7 +40,7 @@ def get_completion(prompt: str, engine: str = "gpt-35-turbo", temperature: float
             )
     return response.choices[0].text
 
-@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
+# @retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
 def get_chat(prompt: str, 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:
     assert model != "text-davinci-003"
     messages = [
@@ -58,3 +59,4 @@ def get_chat(prompt: str, model: str = "gpt-35-turbo", engine: str = "gpt-35-tur
         # request_timeout = 1
     )
     return response.choices[0]["message"]["content"]
+

distillers/guider.py

@@ -107,7 +107,7 @@ class Guidance_Generator():
             reflection_query = self._generate_summary_query(traj, memory[-max_len_mem:])
         else:
             reflection_query = self._generate_summary_query(traj, memory)
-        reflection = get_completion(reflection_query,engine=self.args.gpt_version)
+        reflection = get_completion(reflection_query, engine=self.args.gpt_version)
         logger.info(f'[Reflexion Memory]The reflexion prompt is: {reflection_query}.')
         logger.info(f'[Reflexion Memory]The reflexion response is: {reflection}.')
         return reflection

distillers/traj_prompt_summarizer.py

@@ -54,7 +54,7 @@ class TrajPromptSummarizer():
             reflection_query = self._generate_summary_query(traj, memory[-max_len_mem:])
         else:
             reflection_query = self._generate_summary_query(traj, memory)
-        reflection = get_completion(reflection_query, engine=self.args.gpt_version) 
+        reflection = get_completion(reflection_query, engine=self.args.gpt_version)
         logger.info(f'[Reflexion Memory]The reflexion prompt is: {reflection_query}.')
         logger.info(f'[Reflexion Memory]The reflexion response is: {reflection}.')
         return reflection

main_reflexion.py

@@ -292,8 +292,16 @@ if __name__ == "__main__":
         default=1,
         help="Whether only taking local observations, if is_only_local_obs = 1, only using local obs"
     )
+    parser.add_argument(
+        "--api_type",
+        type=str,
+        default="azure",
+        help="choose api type, now support azure and openai"
+    )
     args = parser.parse_args()
 
+    if args.api_type != "azure" and args.api_type != "openai":
+        raise ValueError(f"The {args.api_type} is not supported, please use 'azure' or 'openai' !")
     # Get the specified translator, environment, and ChatGPT model
     env_class = envs.REGISTRY[args.env]
     init_summarizer = InitSummarizer(envs.REGISTRY[args.init_summarizer], args)