Last active
October 12, 2024 23:54
-
-
Save RohanAwhad/e0c96704f174c88c2352e054be52c519 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| import dataclasses | |
| import openai | |
| import os | |
| import re | |
| import yaml | |
| from pydantic import BaseModel, ValidationError | |
| from typing import Optional | |
| @dataclasses.dataclass | |
| class Message: | |
| role: str | |
| content: str | |
| def llm_call(model: str, messages: list[Message]): | |
| client = openai.OpenAI(api_key=os.environ["TOGETHER_API_KEY"], base_url="https://api.together.xyz/v1") | |
| res = client.chat.completions.create(model=model, messages=[dataclasses.asdict(x) for x in messages], temperature=0.8, max_tokens=4096) | |
| return res.choices[0].message.content | |
| def generate_response_prompt(model: type(BaseModel)) -> str: | |
| TAB = " " | |
| ret = "Respond in YAML format, following the below Pydantic model:\n```python\n" | |
| ret += f"from pydantic import BaseModel, Field\n\nclass {model.__name__}(BaseModel):\n" | |
| yaml_example = "" | |
| for attr, prop in model.model_json_schema()["properties"].items(): | |
| yaml_example += f"\n{attr}: ..." | |
| if 'type' in prop and prop["type"] == "array": | |
| ret += f"{TAB}{attr}: list[{prop['items']['type']}] = Field(desc='{prop['desc']}', max_length={prop['maxItems']})\n" | |
| elif 'anyOf' in prop: | |
| ret += f"{TAB}{attr}: {' | '.join([x['type'] for x in prop['anyOf']])} = Field(desc='{prop['desc']}'" | |
| else: | |
| ret += f"{TAB}{attr}: {prop['type']} = Field(desc='{prop['desc']}')\n" | |
| ret += f"```\n\nYAML should be enclosed in triple backticks like\n```yaml{yaml_example}\n```\n" | |
| return ret | |
| def parse_llm_response(model: type[BaseModel], llm_res: str): | |
| ptrn = re.compile(r"```yaml(.*?)```", re.DOTALL) | |
| match = re.search(ptrn, llm_res) | |
| if match: | |
| try: | |
| res_dict = yaml.safe_load(match.group(1)) | |
| if isinstance(res_dict, dict): ret = model.model_validate(res_dict) | |
| elif isinstance(res_dict, list): ret = [model.model_validate(x) for x in res_dict] | |
| else: raise Exception(f"Shouldn't have reached here. Expected type of dict or list, but got {type(res_dict)}") | |
| return ret | |
| except ValidationError as e: | |
| print("Pydantic Validation Error:", e) | |
| print(llm_res) | |
| except yaml.YAMLError as e: | |
| print("YAML parsing error:", e) | |
| print(llm_res) | |
| except Exception as e: | |
| print("Error:", e) | |
| print(llm_res) | |
| else: | |
| print("Couldn't find yaml tags\nResponse:") | |
| print(llm_res) | |
| return None | |
| def run(model: str, messages: list[Message], max_retries: int, response_model: Optional[type(BaseModel)] = None): | |
| messages[0].content += f"\n---\n\n{generate_response_prompt(response_model)}---\n" | |
| while max_retries: | |
| res = llm_call(model, messages) | |
| ret = parse_llm_response(response_model, res) | |
| if ret is None: max_retries -= 1 | |
| else: return ret |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Usage Example:
Output: