def __init__(self):

# these should be injected but I don't want to change every child class's constructor

config = Config()

self.api : AbstractCompletion = OpenAICompletion(config.get_api_key(), config.get_model())

self.config = config

self.model = config.get_model()

self.printResponse = config.get_print_response()

self.operation = config.get_operation()

self.api_key = config.get_api_key()

self.stream = config.get_stream()

def generate_message_response(self, message, max_tokens=100, temperature=0.5, stream=None):

if(stream is None):

stream = self.stream

if self.printResponse:

print("Modified message: ", message)

if self.operation == 'generate_response':

response = self.api.generate_response(

messages=message,

max_tokens=max_tokens,

n=1,

temperature=temperature,

stream=stream,

)

# TODO: maybe just return the stream? streaming should probably be it's own method

if(stream):

ret = ""

# receive a stream of text from the model

for chunk in response:

if "delta" in chunk.choices[0] and "content" in chunk["choices"][0]["delta"]:

ret += chunk["choices"][0]["delta"]["content"]

print(chunk["choices"][0]["delta"]["content"], end="", flush=True)

print()

# when streaming, don't do any of the extra processing, like saving the response

return ret

if self.printResponse:

print("Complete response: ", response)

return response.choices[0].message.content.strip()

# This option exists for debugging and creating.

elif self.operation == 'count_tokens':

num_tokens = self.num_tokens_from_messages(message)

return num_tokens

else:

raise ValueError(f"Invalid operation: {self.operation}. Supported operations are 'generate_response' and 'count_tokens'.")

def generate_response(self, system_prompt, user_content, max_tokens=100, temperature=0.5, stream=None):

messages = [

{"role": "system", "content": system_prompt},

{"role": "user", "content": user_content},

]

return self.generate_message_response(messages, max_tokens, temperature, stream)

# TODO I never tested these

def num_tokens_from_messages(self, messages, model="gpt-3.5-turbo-0301"):

try:

encoding = tiktoken.encoding_for_model(model)

except KeyError:

encoding = tiktoken.get_encoding("cl100k_base")

if model == "gpt-3.5-turbo-0301":

num_tokens = sum([4 + len(encoding.encode(value)) - 1 if key == "name" else 4 + len(encoding.encode(value)) for message in messages for key, value in message.items()])

num_tokens += 2 * len(messages)

return num_tokens

else:

raise NotImplementedError(f"""num_tokens_from_messages() is not presently implemented for model {model}.

See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.""")

def num_tokens_from_string(self, string, model="gpt-3.5-turbo-0301"):

try:

encoding = tiktoken.encoding_for_model(model)

except KeyError:

encoding = tiktoken.get_encoding("cl100k_base")

if model == "gpt-3.5-turbo-0301":

return 4 + len(encoding.encode(string)) + 2

else:

raise NotImplementedError(f"""num_tokens_from_string() is not presently implemented for model {model}.