Tariq Hasan
Tariq Hasan
I wonder if we should re-run [Unit Test 1.27.1](https://github.com/kubeflow/katib/actions/runs/8840749744/job/24277532658?pr=2316) given that this issue was flagged before in [other PRs](https://github.com/kubeflow/katib/pull/2285#issuecomment-2033391949) and jobs were restarted. Other jobs look to be okay. I...
Hello @andreyvelich I was wondering if I can pick up on this work if you are planning on including this in Katib 0.18.
Sounds good. @andreyvelich I notice that the scope of testing in [training_client_test.py](https://github.com/kubeflow/training-operator/blob/master/sdk/python/kubeflow/training/api/training_client_test.py) is limited to [create_job](https://github.com/kubeflow/training-operator/blob/master/sdk/python/kubeflow/training/api/training_client.py#L306-L485) but that the tests were written before the [train](https://github.com/kubeflow/training-operator/blob/master/sdk/python/kubeflow/training/api/training_client.py#L96-L305) API was implemented. I was...
> > Sounds good. > > I was therefore wondering if we limit the scope of katib_client_test.py to [create_experiment](https://github.com/kubeflow/katib/blob/master/sdk/python/v1beta1/kubeflow/katib/api/katib_client.py#L78-L150) or if we also include the [tune](https://github.com/kubeflow/katib/blob/master/sdk/python/v1beta1/kubeflow/katib/api/katib_client.py#L152-L414) API. > > I...
Are you thinking of adding an example that uses the proposed [tuning API for LLMs](https://github.com/kubeflow/katib/issues/2291) to demonstrate Auto-RAG?
I presume that the initiative here is motivated by the recent trend in the ML space to _fine-tune pre-trained_ models (LLMs or otherwise) using custom datasets instead of _training bare_...
Having worked through the Python SDK and examples for training operator and Katib I have further ideas on an appropriate implementation of the tuning API in Katib for LLMs. It...
> @tariq-hasan LGTM, but the latest commit seems to violate DCO. I addressed the issue. I am wondering if I should squash the commits given that the changes are final...
Makes sense. After reviewing the PR I find that these are the final set of changes - introduce error constants - replace reflect with cmp for error comparison - define...