Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

How to use greedy search correctly #2557

Closed
fan-niu opened this issue Dec 10, 2024 · 4 comments
Closed

How to use greedy search correctly #2557

fan-niu opened this issue Dec 10, 2024 · 4 comments

Comments

@fan-niu
Copy link

fan-niu commented Dec 10, 2024

I used qwen2-1.5b as the basic model to train a machine translation model. I need to use greedy search to generate, but no matter how I set top_p, top_k, and temperature, I cannot get the results of greedy search. Please help me how to set greedy search correctly. Thanks so much !!!

I ran 1000 test sets 2 times with the following configuration, but the results in each round were about 300 different.

  1. top_p=0.0、top_k=0.0、temperature=1.0、beam_width=1
  2. top_p=0.0、top_k=1.0、temperature=0.0、beam_width=1
  3. top_p=1.0、top_k=1.0、temperature=0.0、beam_width=1
  4. top_p=0.0、top_k=1.0、temperature=1.0、beam_width=1

Tensorrtllm version: v0.14.0

@akhoroshev
Copy link
Contributor

try 0.15.0, for me it generates more stable results (compare to 0.14.0)

@fan-niu
Copy link
Author

fan-niu commented Dec 10, 2024

try 0.15.0, for me it generates more stable results (compare to 0.14.0)

@akhoroshev ok got it, thanks.
But still have a question, why 0.14.0 has unstable results on greedy search? thanks

@fan-niu
Copy link
Author

fan-niu commented Dec 11, 2024

@akhoroshev already test it, it's fine, thank you so much !

@nv-guomingz
Copy link
Collaborator

Feel free to reopen this issue if needed.

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

No branches or pull requests

3 participants