Finetuned Version
https://huggingface.co/CyanMonkey/BananaMind-1.5-BB
Hii Mate,
I fine tune your model on my privately cooked dataset.
I evaluated it on Airthmark & get "35.56%"
I hadn't evaluated it on other benchmark though , Please have a look
If You like it then give me some BANANA'🐵🍌
Looks good though i don't really want to train on benchmarks for my main models.
Cool , Yeah
Seems Clearly
A man would not train a 75M param model on 27B tokens for benhcmarks
I would recommend you to go a bit above like in category of sub 150M param
Well what about my Bananas ?
Though Model get worse on inference ig
Cool , Yeah
Seems Clearly
A man would not train a 75M param model on 27B tokens for benhcmarks
I would recommend you to go a bit above like in category of sub 150M paramWell what about my Bananas ?
yes im thinking of a 100M because i want to score higher on the <100M on Open SLM leaderboard, though it will take alot longer because im not paying that high price for the pro 6000 again and on 5070 ti
What about dataset curation?