reasoning_effort
#9
by
owao
- opened
Hello again, bud. Reasoning effort isn't supported right?
passing {"reasoning_effort": "high"} or {"reasoning_effort": "low"} don't seem to make any difference
Thanks by advance and again congrats for the work. I agree with others, it's a great, useful work! That was really needed for this model.
Oh no, it actually makes a difference! Sorry, but that took many attempts before I notice it. It was more blatant for coding tasks.
The generation length contrast between low and high seems less marked than gpt-oss. It still usually thinks for some times even in low. That's interesting.
owao
changed discussion status to
closed
lol)