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

Fix starcoder2 accuracy issue and optimize performance with fused rope #1095

Merged
merged 2 commits into from
Jun 28, 2024

Conversation

mandy-li
Copy link
Collaborator

  1. starcoder2 has accuracy issue when kv_cache is used and max_input_token is specified, such as with this command:

python run_generation.py --model_name_or_path bigcode/starcoder2-3b --batch_size 1 --use_hpu_graphs --use_kv_cache --max_new_tokens 100 --bf16 --prompt "def print_hello_world():" --max_input_tokens 1024

This PR is used to fix the accuracy issue.

  1. Use fused RoPE to optimize the performance on Gaudi

@mandy-li mandy-li requested a review from regisss as a code owner June 25, 2024 05:36
@mandy-li mandy-li added the run-test Run CI for PRs from external contributors label Jun 25, 2024
@HuggingFaceDocBuilderDev

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@regisss regisss merged commit 583037a into main Jun 28, 2024
5 of 6 checks passed
@regisss regisss deleted the mandy/starcoder2 branch June 28, 2024 20:31
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
run-test Run CI for PRs from external contributors
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants