Upgrade openai to >=1.0.0 #1420
Conversation
|
This is a really great PR to see the changes of the new API release, thank you for this. |
etr2460
requested review from
andrew-openai,
jwang47 and
logankilpatrick
as code owners
| if hasattr(openai.error, "set_display_cause"): # type: ignore | ||
| openai.error.set_display_cause() # type: ignore |
There was a problem hiding this comment.
this seems to have been removed in the new version
| @@ -17,7 +17,7 @@ | |||
| ) | |||
|
|
|||
|
|
|||
| @backoff.on_exception(backoff.constant, InvalidRequestError, max_tries=3) | |||
| @backoff.on_exception(backoff.constant, BadRequestError, max_tries=3) | |||
There was a problem hiding this comment.
this seems like it would replicate the old behavior, but the old behavior doesn't seem correct since it's retrying on a bad request which i'd imagine should give the same result on subsequent tries (?)
There was a problem hiding this comment.
Yeah, i agree. I was erring on the side of matching previous behavior vs. fixing anything currently broken.
That said, i can update this to InternalServerError
evals/utils/api_utils.py
Outdated
| openai.APIError, | ||
| openai.RateLimitError, | ||
| openai.APIConnectionError, | ||
| openai.Timeout, |
There was a problem hiding this comment.
could we have someone who's familiar with how the new errors map to the old ones review this part? i think some 4xx shouldn't be retried (e.g. 401), and APIError is raised on both 4xx and 5xx (?)
There was a problem hiding this comment.
APIError, RateLimitError, and and APIConnectionErrors all map 1:1 here.
I think Timeout should actually be replaced with APITimeoutError
ServiceUnavailableError no longer exists, but i'll replace with InternalServerError as mentioned above.
There was a problem hiding this comment.
I've removed APIError as that's a grandparent exception to pretty much all errors from the API. All the others seem reasonable to keep
Migrates evals to the new version of openai-python. Ran the migration script, and then manually fixed issues with running tests/evals Test Plan: - unit tests - run `python -m evals.cli.oaievalset gpt-3.5-turbo test` - test make_me_pay (uses solvers) - run `python -m evals.cli.oaieval langchain/chains/llm_math bigrams --max_samples 20 --dry-run` - run the retrieval-completionfn example
Migrates evals to the new version of openai-python. Ran the migration script, and then manually fixed issues with running tests/evals Test Plan: - unit tests - run `python -m evals.cli.oaievalset gpt-3.5-turbo test` - test make_me_pay (uses solvers) - run `python -m evals.cli.oaieval langchain/chains/llm_math bigrams --max_samples 20 --dry-run` - run the retrieval-completionfn example
Migrates evals to the new version of openai-python. Ran the migration script, and then manually fixed issues with running tests/evals
Test Plan:
python -m evals.cli.oaievalset gpt-3.5-turbo testpython -m evals.cli.oaieval langchain/chains/llm_math bigrams --max_samples 20 --dry-run