Open nqngo opened 5 months ago
Here is my suggestion for default parameters of OpenAI API
frequency_penalty = 0 # penalty for repeating tokens, if set to high -> lower likelihood of repeating word
presence_penalty = 0 # similar to penalty above. Keeping default = 0 for the balance
logprobs = False # whether to return log probabilities in output object
max_tokens = 1000 # maximum number of tokens generated by models
n = 1 # number of response generated by models
seed = 1000 # for reproducible output, must keep this consistent across the codebase
stream = False # whether to stream the output -> display text as stream
temperature = 0.1 # control the creativity of models. Lower values indicate more factual text
ChatGPT is very verbose and provide a fair bit of contextual information on the prompt provide.
When
/ask
is implemented, what are the parameters, token size, words penalty tuning we need to provide to OpenAI to emulate that experience?Please investigate and provide the parameters needed.