[Feature] Local fine-tuning support

[wmpauli]

What feature would you like to see?

Thank you for developing DSPY. It's a great tool. I'm somewhat new to it and am trying to figure out how to perform local fine-tuning with dspy.BootstrapFinetune. Which models are supported? How do I set this up?

On another/related note. I see that dspy.HFClientVLLM has been deprecated. It is mentioned on https://dspy.ai/deep-dive/language_model_clients/lm_local_models/HFClientVLLM/ (w/ a note regarding deprecation), and I'd be curious to know what we should use instead.

Would you like to contribute?

• Yes, I'd like to help implement this.
• No, I just want to request it.

Additional Context

No response

[wmpauli]

Regrading the related note, I believe that this is the answer. Still curious about the local fine-tuning though.

import dspy

model_name = "hosted_vllm/deepseek-ai/deepseek-llm-7b-chat"

vllm_mpt = dspy.LM(
    model_name, api_base="http://localhost:8000/v1/", provider=LocalProvider(), temperature=1.0, cache=False
)
dspy.configure(lm=vllm_mpt)