From 946e15ff2099f662dd7287e720c67f231faa4153 Mon Sep 17 00:00:00 2001 From: Prayag Shah <34435248+prayagnshah@users.noreply.github.com> Date: Sat, 15 Apr 2023 02:22:57 -0300 Subject: [PATCH] Squashed commit of the following: MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit commit 742aaaf9d1ddfee29cc49c993c5d8d0480e53f0a Merge: f2d21e7 fe87faa Author: biobootloader <128252497+biobootloader@users.noreply.github.com> Date: Fri Apr 14 15:44:12 2023 -0700 Merge pull request #13 from fsboehme/main more robust parsing of JSON (+ indentation) commit fe87faa2fb709b782217093ecf88379e153a0f58 Author: Felix Boehme Date: Fri Apr 14 17:49:48 2023 -0400 cleanup commit 4db9d1bf43438a7809e47bd1bd140cad8b3b12e1 Author: Felix Boehme Date: Fri Apr 14 17:49:09 2023 -0400 more cleanup commit e1d0a790f8941a74c4857ac8404c4c1c9e4fb6ed Author: Felix Boehme Date: Fri Apr 14 17:46:18 2023 -0400 cleanup commit b044882dc391d878c570d56ae5e64bb1045d0ec0 Author: Felix Boehme Date: Fri Apr 14 17:37:27 2023 -0400 remove duplicate code from rebase commit dd174cf30eafca66a06e08654854c93ec5297fe0 Author: Felix Boehme Date: Fri Apr 14 17:15:07 2023 -0400 add DEFAULT_MODEL to .env.sample + fix typo commit 2497fb816b862ac1b5d27751c5742fb8171d4207 Author: Felix Boehme Date: Fri Apr 14 16:29:45 2023 -0400 move json_validated_response to standalone function commit 923f7057e36016208f4dbfdf227a0953eba59c47 Author: Felix Boehme Date: Thu Apr 13 11:35:24 2023 -0400 update readme - updated readme to mention .env - added model arg back commit 0656a83da73dd9446406edabdcd607fc5bccc263 Author: Felix Boehme Date: Thu Apr 13 11:29:06 2023 -0400 recursive calls if not json parsable - makes recursive calls to API (with a comment about it not being parsable) if response was not parsable - pass prompt.txt as system prompt - use env var for `DEFAULT_MODEL` - use env var for OPENAI_API_KEY commit 7c072fba2ab3bea728ca5dc3c7b16dc05e3dc54a Author: Felix Boehme Date: Thu Apr 13 11:24:41 2023 -0400 update prompt to make it pay attention to indentation commit c62f91eaeed80501ca3bcc6d15d243bb8aa65a7a Author: Felix Boehme Date: Thu Apr 13 11:23:44 2023 -0400 Update .gitignore commit f2d21e7b93517261161b78acd9fbaf580f6158c1 Merge: 0420860 6343f6f Author: biobootloader <128252497+biobootloader@users.noreply.github.com> Date: Fri Apr 14 13:59:44 2023 -0700 Merge pull request #12 from chriscarrollsmith/main Implemented .env file API key storage commit 6343f6f50be274a49f850426c382cede5652ad62 Author: biobootloader <128252497+biobootloader@users.noreply.github.com> Date: Fri Apr 14 13:59:31 2023 -0700 Apply suggestions from code review commit d87ebfa46f6b830bcfc5c543b149942ce68453cd Merge: 9af5480 75f08e2 Author: Christopher Carroll Smith <75859865+chriscarrollsmith@users.noreply.github.com> Date: Fri Apr 14 16:53:25 2023 -0400 Merge branch 'main' of https://github.com/chriscarrollsmith/wolverine commit 9af5480b89ec58edc11bf8b721eefc137e197d59 Author: Christopher Carroll Smith <75859865+chriscarrollsmith@users.noreply.github.com> Date: Fri Apr 14 16:53:02 2023 -0400 Added python-dotenv to requirements.txt commit 75f08e285293b4482e4fec74a2055629863bc90f Merge: e8a8931 0420860 Author: Christopher Carroll Smith <75859865+chriscarrollsmith@users.noreply.github.com> Date: Fri Apr 14 16:50:29 2023 -0400 Merge pull request #1 from biobootloader/main Reconcile with master branch commit 04208605fe403b70ac5945b4fbcd86e481b8e73d Merge: d547822 6afb4db Author: biobootloader <128252497+biobootloader@users.noreply.github.com> Date: Fri Apr 14 13:22:53 2023 -0700 Merge pull request #20 from eltociear/patch-1 fix typo in README.md commit d54782230c9b30109108e511604c61f8c0c0a001 Merge: 1b9649e 4863df6 Author: biobootloader <128252497+biobootloader@users.noreply.github.com> Date: Fri Apr 14 13:19:43 2023 -0700 Merge pull request #17 from hemangjoshi37a/main added `star-history` ⭐⭐⭐⭐⭐ commit 6afb4db2ffc7878e2a125cd53917a1abfacb8790 Author: Ikko Eltociear Ashimine Date: Fri Apr 14 16:37:05 2023 +0900 fix typo in README.md reliablity -> reliability commit 4863df689877d0628520a10346c6ad7cbb7cd9cd Author: Hemang Joshi Date: Fri Apr 14 10:27:32 2023 +0530 added `star-history` added `star-history` commit e8a893156e097d8c964d0a8ff195b00fdb536fad Author: Christopher Carroll Smith <75859865+chriscarrollsmith@users.noreply.github.com> Date: Wed Apr 12 13:45:54 2023 -0400 Implemented .env file API key storage --- .env.sample | 2 + .gitignore | 5 ++- README.md | 15 +++++++- prompt.txt | 5 ++- requirements.txt | 1 + wolverine.py | 97 ++++++++++++++++++++++++++++++++++-------------- 6 files changed, 94 insertions(+), 31 deletions(-) create mode 100644 .env.sample diff --git a/.env.sample b/.env.sample new file mode 100644 index 0000000..2184c45 --- /dev/null +++ b/.env.sample @@ -0,0 +1,2 @@ +OPENAI_API_KEY=your_api_key +#DEFAULT_MODEL=gpt-3.5-turbo diff --git a/.gitignore b/.gitignore index 673fe09..da34ea7 100644 --- a/.gitignore +++ b/.gitignore @@ -1,2 +1,5 @@ venv -openai_key.txt +.venv +.env +env/ +.vscode/ diff --git a/README.md b/README.md index c988276..98571ca 100644 --- a/README.md +++ b/README.md @@ -13,8 +13,11 @@ For a quick demonstration see my [demo video on twitter](https://twitter.com/bio python3 -m venv venv source venv/bin/activate pip install -r requirements.txt + cp .env.sample .env -Add your openAI api key to `openai_key.txt` - _warning!_ by default this uses GPT-4 and may make many repeated calls to the api. +Add your openAI api key to `.env` + +_warning!_ By default wolverine uses GPT-4 and may make many repeated calls to the api. ## Example Usage @@ -30,13 +33,21 @@ You can also use flag `--confirm=True` which will ask you `yes or no` before mak python wolverine.py buggy_script.py "subtract" 20 3 --confirm=True +If you want to use GPT-3.5 by default instead of GPT-4 uncomment the default model line in `.env`: + + DEFAULT_MODEL=gpt-3.5-turbo + ## Future Plans This is just a quick prototype I threw together in a few hours. There are many possible extensions and contributions are welcome: - add flags to customize usage, such as asking for user confirmation before running changed code - further iterations on the edit format that GPT responds in. Currently it struggles a bit with indentation, but I'm sure that can be improved -- a suite of example buggy files that we can test prompts on to ensure reliablity and measure improvement +- a suite of example buggy files that we can test prompts on to ensure reliability and measure improvement - multiple files / codebases: send GPT everything that appears in the stacktrace - graceful handling of large files - should we just send GPT relevant classes / functions? - extension to languages other than python + +## Star History + +[![Star History Chart](https://api.star-history.com/svg?repos=biobootloader/wolverine&type=Date)](https://star-history.com/#biobootloader/wolverine) diff --git a/prompt.txt b/prompt.txt index 4376ab2..ec0582a 100644 --- a/prompt.txt +++ b/prompt.txt @@ -4,10 +4,13 @@ Because you are part of an automated system, the format you respond in is very s In addition to the changes, please also provide short explanations of the what went wrong. A single explanation is required, but if you think it's helpful, feel free to provide more explanations for groups of more complicated changes. Be careful to use proper indentation and spacing in your changes. An example response could be: +Be ABSOLUTELY SURE to include the CORRECT INDENTATION when making replacements. + +example response: [ {"explanation": "this is just an example, this would usually be a brief explanation of what went wrong"}, {"operation": "InsertAfter", "line": 10, "content": "x = 1\ny = 2\nz = x * y"}, {"operation": "Delete", "line": 15, "content": ""}, - {"operation": "Replace", "line": 18, "content": "x += 1"}, + {"operation": "Replace", "line": 18, "content": " x += 1"}, {"operation": "Delete", "line": 20, "content": ""} ] diff --git a/requirements.txt b/requirements.txt index 9d61d58..3923778 100644 --- a/requirements.txt +++ b/requirements.txt @@ -13,6 +13,7 @@ multidict==6.0.4 openai==0.27.2 pycodestyle==2.10.0 pyflakes==3.0.1 +python-dotenv==1.0.0 requests==2.28.2 six==1.16.0 termcolor==2.2.0 diff --git a/wolverine.py b/wolverine.py index 42001b6..363e3da 100644 --- a/wolverine.py +++ b/wolverine.py @@ -5,13 +5,20 @@ import os import shutil import subprocess import sys - import openai from termcolor import cprint +from dotenv import load_dotenv + # Set up the OpenAI API -with open("openai_key.txt") as f: - openai.api_key = f.read().strip() +load_dotenv() +openai.api_key = os.getenv("OPENAI_API_KEY") + +DEFAULT_MODEL = os.environ.get("DEFAULT_MODEL", "gpt-4") + + +with open("prompt.txt") as f: + SYSTEM_PROMPT = f.read() def run_script(script_name, script_args): @@ -25,7 +32,49 @@ def run_script(script_name, script_args): return result.decode("utf-8"), 0 -def send_error_to_gpt(file_path, args, error_message, model): +def json_validated_response(model, messages): + """ + This function is needed because the API can return a non-json response. + This will run recursively until a valid json response is returned. + todo: might want to stop after a certain number of retries + """ + response = openai.ChatCompletion.create( + model=model, + messages=messages, + temperature=0.5, + ) + messages.append(response.choices[0].message) + content = response.choices[0].message.content + # see if json can be parsed + try: + json_start_index = content.index( + "[" + ) # find the starting position of the JSON data + json_data = content[ + json_start_index: + ] # extract the JSON data from the response string + json_response = json.loads(json_data) + except (json.decoder.JSONDecodeError, ValueError) as e: + cprint(f"{e}. Re-running the query.", "red") + # debug + cprint(f"\nGPT RESPONSE:\n\n{content}\n\n", "yellow") + # append a user message that says the json is invalid + messages.append( + { + "role": "user", + "content": "Your response could not be parsed by json.loads. Please restate your last message as pure JSON.", + } + ) + # rerun the api call + return json_validated_response(model, messages) + except Exception as e: + cprint(f"Unknown error: {e}", "red") + cprint(f"\nGPT RESPONSE:\n\n{content}\n\n", "yellow") + raise e + return json_response + + +def send_error_to_gpt(file_path, args, error_message, model=DEFAULT_MODEL): with open(file_path, "r") as f: file_lines = f.readlines() @@ -34,12 +83,7 @@ def send_error_to_gpt(file_path, args, error_message, model): file_with_lines.append(str(i + 1) + ": " + line) file_with_lines = "".join(file_with_lines) - with open("prompt.txt") as f: - initial_prompt_text = f.read() - prompt = ( - initial_prompt_text + - "\n\n" "Here is the script that needs fixing:\n\n" f"{file_with_lines}\n\n" "Here are the arguments it was provided:\n\n" @@ -51,28 +95,27 @@ def send_error_to_gpt(file_path, args, error_message, model): ) # print(prompt) + messages = [ + { + "role": "system", + "content": SYSTEM_PROMPT, + }, + { + "role": "user", + "content": prompt, + }, + ] - response = openai.ChatCompletion.create( - model=model, - messages=[ - { - "role": "user", - "content": prompt, - } - ], - temperature=1.0, - ) - - return response.choices[0].message.content.strip() + return json_validated_response(model, messages) -# Added the flag confirm. Once user use flag confirm then it will ask for confirmation before applying the changes. -def apply_changes(file_path, changes_json, confirm=False): +def apply_changes(file_path, changes: list, confirm=False): + """ + Pass changes as loaded json (list of dicts) + """ with open(file_path, "r") as f: original_file_lines = f.readlines() - changes = json.loads(changes_json) - # Filter out explanation elements operation_changes = [change for change in changes if "operation" in change] explanations = [ @@ -137,8 +180,7 @@ def apply_changes(file_path, changes_json, confirm=False): print("Changes applied.") -# Added the flag confirm. Once user use flag confirm then it will ask for confirmation before applying the changes. -def main(script_name, *script_args, revert=False, model="gpt-4", confirm=False): +def main(script_name, *script_args, revert=False, model=DEFAULT_MODEL, confirm=False): if revert: backup_file = script_name + ".bak" if os.path.exists(backup_file): @@ -169,6 +211,7 @@ def main(script_name, *script_args, revert=False, model="gpt-4", confirm=False): error_message=output, model=model, ) + apply_changes(script_name, json_response, confirm=confirm) cprint("Changes applied. Rerunning...", "blue")