Replit-v2-CodeInstruct vs Replit-v1-CodeInstruct
#5
Unanswered
matthiasgeihs
asked this question in
Q&A
Replies: 1 comment 1 reply
-
|
Hi, v1 model is trained on 512 sequence length. V2 is trained on 2000 sequence length. The longer sequence training from what I know wasn’t as effective because of the training data used |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
Yes, I've seen this, but it shouldn't make the results worse? |
Beta Was this translation helpful? Give feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Why does
Replit-v2-CodeInstructperform significantly worse thanReplit-v1-CodeInstruct?(21.5% vs 25.8% pass@1)
Beta Was this translation helpful? Give feedback.
All reactions