diff --git a/pylib/llm_wrapper.py b/pylib/llm_wrapper.py index 6c97cf7..c449e1b 100644 --- a/pylib/llm_wrapper.py +++ b/pylib/llm_wrapper.py @@ -12,6 +12,7 @@ ''' import os +import json import asyncio import concurrent.futures from functools import partial @@ -70,8 +71,14 @@ def from_openai_chat(response): if 'message' in c: c['message'] = llm_response(c['message']) rc1 = resp['choices'][0] - # print(f'from_openai_chat: {rc1 =}') - resp['first_choice_text'] = rc1['text'] if 'text' in rc1 else rc1['message']['content'] + # No response message content if a tool call is invoked + if 'tool_calls' in rc1['message']: + # Why the hell does OpenAI have these arguments properties as plain text? Seems like a massive layering violation + for tc in rc1['message']['tool_calls']: + tc['function']['arguments_obj'] = json.loads(tc['function']['arguments']) + else: + resp['first_choice_text'] = rc1['text'] if 'text' in rc1 else rc1['message']['content'] + print('GRIPPO', f'from_openai_chat: {rc1 =}') else: resp['first_choice_text'] = resp['content'] return resp @@ -244,19 +251,6 @@ def available_models(self) -> List[str]: raise RuntimeError(f'Unexpected response from {self.base_url}/models:\n{repr(resp)}') return [ i['id'] for i in resp['data'] ] - @staticmethod - def first_choice_text(response): - ''' - Given an OpenAI-compatible API simple completion response, return the first choice text - ''' - warnings.warn('The first_choice_text method is deprecated; use the first_choice_text attribute or key instead', DeprecationWarning, stacklevel=2) # noqa E501 - try: - return response.choices[0].text - except AttributeError: - raise RuntimeError( - f'''Response does not appear to be an OpenAI API completion structure, as expected: -{repr(response)}''') - class openai_chat_api(openai_api): ''' @@ -322,19 +316,6 @@ async def __call__(self, prompt, api_func=None, **kwargs): # Haven't implemented any OpenAI API calls that are async, so just call the sync version return self.call(prompt, api_func, **kwargs) - @staticmethod - def first_choice_message(response): - ''' - Given an OpenAI-compatible API chat completion response, return the first choice message content - ''' - warnings.warn('The first_choice_message method is deprecated; use the first_choice_text attribute or key instead', DeprecationWarning, stacklevel=2) # noqa E501 - try: - return response.choices[0].message.content - except AttributeError: - raise RuntimeError( - f'''Response does not appear to be an OpenAI API chat-style completion structure, as expected: -{repr(response)}''') - class llama_cpp_http(llm_wrapper): ''' @@ -465,19 +446,6 @@ async def __call__(self, messages, req='/v1/chat/completions', timeout=30.0, api else: raise RuntimeError(f'Unexpected response from {self.base_url}{req}:\n{repr(result)}') - @staticmethod - def first_choice_message(response): - ''' - Given an OpenAI-compatible API chat completion response, return the first choice message content - ''' - warnings.warn('The first_choice_message method is deprecated; use the first_choice_text attribute or key instead', DeprecationWarning, stacklevel=2) # noqa E501 - try: - return response['choices'][0]['message']['content'] - except (IndexError, KeyError): - raise RuntimeError( - f'''Response does not appear to be a llama.cpp API chat-style completion structure, as expected: -{repr(response)}''') - class ctransformer: '''