Will the codex openai library work with fauxpilot?
smith-co opened this issue · comments
smith-coding commented
Thanks for this fantastic work. I will do a quick comparison for code completion between fauxpiot and copilot.
I will be making calls with the following API invocation:
response = openai.Completion.create(
model="code-davinci-002",
prompt=input_prompt,
temperature=temperature,
max_tokens=max_tokens,
top_p=1,
n=number_of_suggestions,
frequency_penalty=frequency_penalty,
presence_penalty=0,
stop="###",
)
# suggestions = response['choices']
result = ""
if 'choices' in response:
x = response['choices']
if len(x) > 0:
for i in range(0, len(x)):
result = x[i]['text']
else:
result = ''
# are these metrics present?
response_completion_tokens = response["usage"]["completion_tokens"]
response_prompt_tokens = response["usage"]["prompt_tokens"]
response_total_tokens = response["usage"]["total_tokens"]
Can you please let me know whether this API invocation would work for fauxpilot?
Brendan Dolan-Gavitt commented
The usage metrics are computed and returned by FauxPilot, so yes, this should work. One thing to be aware of is that the context length supported by code-davinci-002
is 8000, whereas the CodeGen models only support up to 2048, so you won't be able to directly compare prompts longer than that.
Brendan Dolan-Gavitt commented
I'll close this for now but let me know if you have other questions!