reduced multi language quality
#11
by
rastegar
- opened
Hi, I tested this model over multi language tasks like translation, quality reduced too much compared to 235B instruct, It impacted higher in languages like persian or arabic
compared to 235B instruct
Well, it's a 80B-A3B model... What did you expect? The real question is whether it's better or worse than 30B-A3B.
Honestly, the only thing blocking me from launching this 30B model into production is the grammatical weaknesses—in my case, the Polish language. The rest of the model's performance is top-notch.