You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I was wondering whether which version of CodeGemma 7B is used in the papers. In HuggingFace there are two versions, the Base model and the Instruct model. In NIM there is only one version of CodeGemma 7B but it is not specified if it is the base or the instruct model.
I ask this because I am having a hard time replicating the results locally using HF models with the base model alone, the instruct model seems more aligned with your paper (7% vs 12% on Code Completition and 9% vs 11 on Spec to RTL).
The text was updated successfully, but these errors were encountered:
I was wondering whether which version of CodeGemma 7B is used in the papers. In HuggingFace there are two versions, the Base model and the Instruct model. In NIM there is only one version of CodeGemma 7B but it is not specified if it is the base or the instruct model.
I ask this because I am having a hard time replicating the results locally using HF models with the base model alone, the instruct model seems more aligned with your paper (7% vs 12% on Code Completition and 9% vs 11 on Spec to RTL).
The text was updated successfully, but these errors were encountered: