test: reduce code redundancy in openai based tests #54
Conversation
codefromthecrypt
force-pushed
the
refactor-tests
branch
from
667296c
to
ca489fc
Compare
codefromthecrypt
changed the title
|
@michaelneale I scrubbed out the model change. this branch helps me in testing azure in another PR. PTAL? |
|
I think need to pull in change to make it more deterministic |
Signed-off-by: Adrian Cole <adrian.cole@elastic.co>
codefromthecrypt
force-pushed
the
refactor-tests
branch
from
ca489fc
to
f8044b7
Compare
|
PTAL I also made a small internal change which allows me to in a subsequent PR re-use the openai impl for azure |
|
brought in some generic changes from the dependent PR #60 |
Signed-off-by: Adrian Cole <adrian.cole@elastic.co>
codefromthecrypt
force-pushed
the
refactor-tests
branch
from
fc8ab2c
to
c5dc482
Compare
Signed-off-by: Adrian Cole <adrian.cole@elastic.co>
codefromthecrypt
force-pushed
the
refactor-tests
branch
from
2f972bd
to
095ecb3
Compare
There was a problem hiding this comment.
looks ok to me - @baxen probably needs blessing as touches openai provider - just as extra eyeballs
| print("Completion content from OpenAI:", reply[0].content) | ||
|
|
||
|
|
||
| @pytest.mark.vcr() |
There was a problem hiding this comment.
while these tests down look redundant to the ones in the parent dir, what's nice is the non_integration ones are using recorded requests. so we can see exactly what should be happening. This has been helpful figuring out what should work in azure openai, because it works in normal openai, but doesn't
| client.get("") | ||
| # The OpenAI API is defined after "v1/", so we need to join it here. | ||
| client.base_url = client.base_url.join("v1/") |
There was a problem hiding this comment.
nit: i'd put this above on line 37
There was a problem hiding this comment.
ill do a quick fix in the 0.9.3 release so we can get this in
There was a problem hiding this comment.
will do on next change thanks!
There was a problem hiding this comment.
oh it is subtle. the health check is ollama, so we need to test the base of ollama before appending "v1". It would be better if we didn't need to health check imho, especially as we aren't in any other provider. lemme know your thoughts on that.
There was a problem hiding this comment.
found a way that may work fine in #60
This reduces fixture redundancy in openai by sharing functions. This also clarifies the openai base endpoint for future consolidation with azure.
note: in order to be able to run just integration -k openai and not also run ollama, I needed to get rid of the openai subdir in the tests (as it was before I started)