Skip to content

Conversation

@epicfilemcnulty
Copy link
Contributor

min_p is a popular sampling method that has been adopted by various backends, e.g exllama, llamacpp, vllm.

@albertfgu
Copy link
Contributor

Thank you! Can you provide an example generation command for testing?

@epicfilemcnulty
Copy link
Contributor Author

@albertfgu Well, it's just a regular call to model.generate, e.g.

 output_ids = model.generate(
        input_ids=input_ids,
        max_length=256,
        temperature=0.8,
        top_p=1,
        top_k=0,
        min_p=0.05
 )

Or shall I also modify benchmark_generation_mamba_simple.py to include min_p option?

@albertfgu
Copy link
Contributor

Yes, thank you, I just wanted to test that there's a simple command that can be run that produces reasonable outputs.

@albertfgu albertfgu merged commit c7bca02 into state-spaces:main Jan 30, 2024
@hp-l33 hp-l33 mentioned this pull request Aug 3, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants