context size / sampling params / reasoning
#1
by
owao
- opened
Hey! Thank you really much for putting it out there! And thanks for the effort to make it available through llama.cpp!
Some questions:
- is the context window 131072 tokens natively without rope scaling?
- what value would you recommend for max_new_tokens?
- what sampling params are important and what recommended values?
- is this a non-reasoning-only model? I didn't find any special tokens like
<think>or similar, but the name of the model confused me.
Thanks by advance!
owao
changed discussion title from
context size and sampling params
to context size / sampling params / reasoning
No updates?