Model Tuning broken and requesting to support more Tasks

[adfayed]

Context

What are you trying to do and how would you want to do it differently? Is it something you currently you cannot do? Is this related to an issue/problem?

Answer: Trying to model tune for chat purpose. Currently not possible. Also related issue/problem, tuning is currently broken since using a validation file is commented out. Tuning needs to take into account validation.

---

Tuning works for classification and summarization tasks, but not for chat:

        tune_result = self.client.tune.create(
            model_id=model_id,
            name=f"{model_id}-tuned",
            tuning_type=TuningType.PROMPT_TUNING,
            task_id="classification",  # Another supported task is "summarization"
            parameters=hyperparams,
            training_file_ids=[training_file_id]
        ).result

I think it would be super useful to get the chat task supported as well. That is probably the highest use-case for other folks besides myself. Also a lot of the top performing models only support chat and generation like mistralai/mixtral-8x7b-instruct-v01 and meta-llama/llama-3-70b-instruct.

Alternatives

Can you achieve the same result doing it in an alternative way? Is the alternative considerable?

Answer: Not really, the top performing models that I'd like to use (mentioned above) do not support classification nor summarization.

Has the feature been requested before?

Answer: No

If the feature request is approved, would you be willing to submit a PR?

Answer: Yes, sure thing! I'd just need some guidance on where to get started.

[adfayed]

Any updates on this please?

[RacketyWater7]

Hi guys, is there any update on this issue yet?