Wowowowow
What's happening here
@bartowski
?
Did they release a new version with further RL without even teasing anything??
I saw in addition to reuploading the safetensors, they also added an incredible comparative chart.
I always had this QWQ in my heart :D QWQ, marco-o1, Alibaba were the pioneers!
So: is it a massive comeback!!?
I didn't find new info in their blog post.
OH ok!!! I just saw it was the final model! I almost forgot the first one was a preview!
I'm so looking to try this! They rock!
and thanks for the quants as always <3
Haha yes this is a very exciting release! They've been teasing it about a week now, and their preview back in January was quite promising, super awesome from them!
Already testing it and it's great. Just with some prompts it's thinking for ages...And because it's so slow on my hardware, it lasts forever. One of the promts to make it think forever:) >>>
Write a Python function that prints the next 20 leap years. Reply with only the function.
Already testing it and it's great. Just with some prompts it's thinking for ages...And because it's so slow on my hardware, it lasts forever. One of the promts to make it think forever:) >>>
Write a Python function that prints the next 20 leap years. Reply with only the function.
ah! I feel you! When you start using thinking models, you end up spending far more time refining your prompts :D And I think that's a healthier approach than throwing a dirty prompt at it and hitting generation 5 times in a row. Thinking models make you more resource conscious!
That said, I wouldn't have expected this task to use thinking a lot, but I guess more thinking is never too much :D
I have 30 usual test questions, and I'm impressed so far. First time EVER i got correct answer to this question:
Tell me the name of a country whose name ends with 'lia'. Give me the capital city of that country as well.
Answer: Australia