Benchmarks
#4
by
olisicky
- opened
Hi! Thank you for sharing the model. Have you tried any benchmarks after the continuous pre-training? I found similar work just for chinese with Llama3 where they report a small increase in e.g., CMMLU compared to the original Llama3-8B. On the other hand, the drop on English MMLU is quite large. We are trying continuous pre-training for Czech language but the results show similar behavior. It would be great if you could share some insights! Thank you very much!
Ondřej
currently this model is under fix, so new models are coming soon, so after then I could upload test with some benchmarks :)
beomi
changed discussion status to
closed