linux_test - #1337
Conversation
|
@makelinux is this a duplicate of #1314? Adding |
|
@nathan-weinberg , no, it is another better method |
|
@makelinux in the future, please limit changes to a single PR - if you are experimenting with implementations, you can open a draft PR and open it for review once it is ready |
|
This pull request has merge conflicts that must be resolved before it can be |
|
This pull request has merge conflicts that must be resolved before it can be |
|
This pull request has merge conflicts that must be resolved before it can be |
|
This pull request has merge conflicts that must be resolved before it can be |
Prepare for linux_test(). Connection to local server for testing is possible without TLS. Signed-off-by: Costa Shulyupin <costa.shul@redhat.com>
No code changes, just change in indentation. Preparation for linux_test(). Signed-off-by: Costa Shulyupin <costa.shul@redhat.com>
linux_test is implemented like the chat and the generate with OpenAI client. Only `ensure_server` and starting the server defines `model_path`, client.chat.completions.create ignores it. Don't use TLS for simplicity, assuming it's a local server only because fine-tuned model is local. Resolves #659 Signed-off-by: Costa Shulyupin <costa.shul@redhat.com>
test_model_train() emulates very heavy `ilab model train' just to make work for `test_model_test()`. Signed-off-by: Costa Shulyupin <costa.shul@redhat.com>
RobotSail
left a comment
There was a problem hiding this comment.
Thank you for your contribution, LGTM!
| from .vllm import Server as vllm_server | ||
|
|
||
| model_path = pathlib.Path(cfg.model_path) | ||
| model_path = pathlib.Path(model_path or cfg.model_path) |
There was a problem hiding this comment.
I'm not going to block on this but just as an FYI you should prefer to use cfg.model_path if not model_path else model_path in python when doing these default evaluations.
Already supported, see instructlab#1337 Signed-off-by: Costa Shulyupin <costa.shul@redhat.com>
linux_test is implemented like the chat and the generate with OpenAI client.
Don't use TLS for simplicity, assuming it's a local server only because fine-tuned model is local.
Resolves #659
Testing: #632
Checklist:
conventional commits.