Merge pull request #9 from mail-ecnu/master

README.md

@@ -34,14 +34,16 @@ For `L5` level, we handcraft the few shot examples with domain knowledge in `pro
 ```python 
 import openai
 class gpt: 
-    def __init__(self,):
-        openai.api_type = "azure"
-        openai.api_version = "2023-05-15"
-        # Your Azure OpenAI resource's endpoint value.
-        openai.api_base = "https://js-partner.openai.azure.com/"
-        openai.api_key = "your azure openai key"
+    def __init__(self, args):
+        if args.api_type == "azure":
+            openai.api_type = "azure"
+            openai.api_version = "2023-05-15"
+            # Your Azure OpenAI resource's endpoint value.
+            openai.api_base = "https://midivi-main-scu1.openai.azure.com/"
+            openai.api_key = "your azure key"
+        else:
+            openai.api_key = "your openai key"
 ```
-
 2. Install Requirements
 
 ```
@@ -49,13 +51,19 @@ conda env create --file environment.yaml
 ```
 
 3. Testing 
-The project can be run using the provided test.sh script. This script runs a series of commands, each of which initiates a Gym environment and applies different translators to it.
+The project can be run using the provided .sh script in shell/ folder. This script runs a series of commands, each of which initiates a Gym environment and applies different translators to it.
 
 Here is an example of how to run the script:
 
 ```
-./test.sh
+sh shell/test_cartpole.sh
+```
+Or you can also test this by copying a command from a .sh script
 ```
+python main_reflexion.py --env_name CartPole-v0 --init_summarizer cart_init_translator --curr_summarizer cart_basic_translator --decider exe_actor --prompt_level 1 --num_trails 1 --distiller guide_generator 
+```
+
+If you use openai key, please add "--api_type openai" at the end of the command!
 
 ### Install Mujoco Environment
 1. Download the MuJoCo, recommand [mujoco210](https://github.com/google-deepmind/mujoco/releases/tag/2.1.0), for Linux, it is `mujoco210-linux-x86_64.tar.gz

deciders/act.py

@@ -34,7 +34,7 @@ class NaiveAct(gpt):
         else:
             model = args.gpt_version
         self.encoding = tiktoken.encoding_for_model(model)
-        super().__init__()
+        super().__init__(args)
         self.distiller = distiller
         self.fewshot_example_initialization(args.prompt_level, args.prompt_path, distiller = self.distiller)
         self.default_action = 1

deciders/utils.py

@@ -16,8 +16,8 @@ else:
 
 Model = Literal["gpt-4", "gpt-35-turbo", "text-davinci-003"]
 
-from .gpt import gpt
-gpt().__init__()
+# from .gpt import gpt
+# gpt().__init__()
 
 import timeout_decorator
 @timeout_decorator.timeout(30)

environment.yaml

@@ -86,6 +86,7 @@ dependencies:
   - zeromq=4.3.4
   - zlib=1.2.13
   - pip:
+      - ale-py==0.8.1
       - absl-py==1.4.0
       - aiohttp==3.8.4
       - aiosignal==1.3.1
@@ -185,3 +186,4 @@ dependencies:
       - win32-setctime==1.1.0
       - yarl==1.9.2
       - zipp==3.15.0
+      - git+ssh://[email protected]/hyyh28/atari-representation-learning.git

main_reflexion.py

@@ -303,6 +303,15 @@ if __name__ == "__main__":
 
     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' !")
+    
+    # Please note when using "azure", the model name is gpt-35-turbo while using "openai", the model name is "gpt-3.5-turbo"
+    if args.api_type == "azure":
+        if args.gpt_version == "gpt-3.5-turbo":
+            args.gpt_version = 'gpt-35-turbo'
+    elif args.api_type == "openai":
+        if args.gpt_version == "gpt-35-turbo":
+            args.gpt_version = 'gpt-3.5-turbo'
+
     # Get the specified translator, environment, and ChatGPT model
     env_class = envs.REGISTRY[args.env]
     init_summarizer = InitSummarizer(envs.REGISTRY[args.init_summarizer], args)