Add detailed Readme and Dockerfile

Signed-off-by: Sebastiano Mariani <[email protected]>

Author: Phat3

Dockerfile

@@ -0,0 +1,13 @@
+FROM python:3.11
+
+WORKDIR /app
+
+# Copy the requirements file
+COPY ./src /app
+COPY ./pyproject.toml /app
+
+# Install the requirements
+RUN pip install -e .
+
+# Run the app
+ENTRYPOINT ["llm-repl"]

README.md

@@ -1,17 +1,87 @@
 # LLM REPL
 
-This repository contains a REPL interface for interacting with various LLMs such as the ChatGPT, etc.
+## What is this?
+
+The goal of this project is to create a simple, interactive **REPL** (Read-Eval-Print-Loop) that allows users to interact with a variety of Large Language Models (**LLMs**). The project is mainly built on top of two Python libraries: [langchain](https://github.com/hwchase17/langchain), which provides a convenient and flexible interface for working with LLMs, and [rich](https://github.com/Textualize/rich) which provides a user-friendly interface for the REPL.
+
+Currently, the project is in development and only supports interaction with the ChatGPT but it has been structure to make it easy to extend it use any LLMs, including custom ones (by extending `BaseLLM` in `./src/llm_repl/llms/__init__.py`).
+
+ChatGPT can be interacted by using the models `gpt-3.5-turbo` and `gpt4` (For users who got GPT-4 API beta).
+
+## Features
+
+The REPL supports the following features:
+
+### Streaming Mode
+
+The REPL won't wait for the model to finish generating the output, but it will start printing the output as soon as it is available.
+
+### Pretty Printing
+
+The REPL supports Markdown rendering both of the input and the output.
+
+PS: In this initial version of the REPL, the full Markdown syntax is only when running the tool in `non-streaming` mode. In `streaming` mode only code sections will be pretty printed.
+
+### Model Switching
+
+The REPL supports the switching between different models. At the moment, the only supported LLMs are `chatgpt` and `chatgpt4`.
 
 ## Installation
 
-To install the TUI in development mode, first setup the `pre-commit` hooks:
+```bash
+pip install llm-repl
+```
+
+## Usage
+
+First export your OpenAI API key as an environment variable:
 
 ```bash
-pre-commit install
+export OPENAI_API_KEY=<OPENAI_KEY>
+```
+
+Then run the REPL:
+
+```bash
+llm-repl
+```
+
+Or if you want to use a specific model:
+
+```bash
+llm-repl --llm chatgpt4
+```
+
+### Run inside Docker
+
+```bash
+docker run -it --rm -e OPENAI_API_KEY=<OPENAI_KEY> phate/llm-repl
+```
+
+Or if you want to source the environment variables from a file, first create a file called `.env` with the following content:
+
+```bash
+OPENAI_API_KEY=<OPENAI_KEY>
 ```
 
+And then run the following command:
+
+```bash
+docker run -it --rm --env-file .env phate/llm-repl
+```
+
+## Development
+
+To install the REPL in development mode
+
 Then install the package in development mode:
 
 ```bash
 pip install -e ".[DEV]"
 ```
+
+Before contributing, please make sure to run the following commands:
+
+```bash
+pre-commit install
+```

src/llm_repl/__main__.py

@@ -15,7 +15,7 @@
 def main():
     parser = argparse.ArgumentParser(description="LLM REPL")
     parser.add_argument(
-        "--model",
+        "--llm",
         type=str,
         default="chatgpt",
         help="The LLM model to use",
@@ -32,4 +32,4 @@ def _(event):
         repl.handle_enter(event)
 
     # Run the REPL
-    repl.run(MODELS[args.model])
+    repl.run(MODELS[args.llm])

src/llm_repl/repl.py

@@ -14,11 +14,12 @@
 
 class LLMRepl:
 
-    INTRO_BANNER = "Welcome to LLM REPL! Input your message and press enter twice to send it to the LLM (Ctrl+C to exit)"
     LOADING_MSG = "Thinking..."
     SERVER_MSG_TITLE = "LLM"
     CLIENT_MSG_TITLE = "You"
     ERROR_MSG_TITLE = "ERROR"
+    EXIT_TOKEN = "exit"
+    INTRO_BANNER = f"Welcome to LLM REPL! Input your message and press enter twice to send it to the LLM (type '{EXIT_TOKEN}' to quit the application)"
 
     def __init__(self, config: dict[str, Any]):
         self.console = Console()
@@ -35,7 +36,7 @@ def __init__(self, config: dict[str, Any]):
         self.server_color = config["style"]["server"]["color"]
         self.error_color = "bold red"
         self.misc_color = "gray"
-        self.model = None  # Optional[BaseLLM] = None
+        self.llm = None  # Optional[BaseLLM] = None
 
     def handle_enter(self, event):
         """
@@ -104,7 +105,7 @@ def print_misc_msg(self, msg: str, justify: str = "left"):
         """
         self._print_msg("", msg, self.misc_color, justify=justify)
 
-    def run(self, model):
+    def run(self, llm):
         """
         Starts the REPL.
 
@@ -113,31 +114,35 @@ def run(self, model):
         The user can enter new lines in the REPL by pressing Enter once. The
         REPL will terminate when the user presses Enter twice.
 
-        :param BaseLLM model: The LLM model to use.
+        :param BaseLLM llm: The LLM to use.
         """
 
-        self.model = model.load(self)
-        if self.model is None:
+        self.llm = llm.load(self)
+        if self.llm is None:
             return
 
         self.print_misc_msg(self.INTRO_BANNER, justify="center")
-        self.print_misc_msg(f"Loaded model: {self.model.name}", justify="center")
+        self.print_misc_msg(f"Loaded model: {self.llm.name}", justify="center")
 
         while True:
             user_input = self.session.prompt("> ").rstrip()
+            if user_input == self.EXIT_TOKEN:
+                self.print_misc_msg("Bye!")
+                break
+
             self.print_client_msg(user_input)
 
-            if not self.model.is_in_streaming_mode:
+            if not self.llm.is_in_streaming_mode:
                 self.print_misc_msg(self.LOADING_MSG)
             else:
                 self.console.rule(
                     f"[{self.server_color}]{self.SERVER_MSG_TITLE}",
                     style=self.server_color,
                 )
 
-            resp = self.model.process(user_input)
+            resp = self.llm.process(user_input)
 
-            if not self.model.is_in_streaming_mode:
+            if not self.llm.is_in_streaming_mode:
                 self.print_server_msg(resp)
             else:
                 self.console.print()