Model speed

#1
by mduran159 - opened

the meta-llama/Llama-3.2-1B-Instruct is even faster than this model. Weird thing, because last version for 3.1 was great. I will be waiting for the fix, for sure.

mduran159 changed discussion status to closed

nevermind Im an Idiot xd

mduran159 changed discussion status to open

Well, now comparing, Its almost at the same speed than a Llama-3.2-3B-Instruct. A model with 2B more of parameters and this model is supossed to be faster. Maybe I'm still being an idiot and I'm missing something, but I think now models are in equal conditions and unsloth/Llama-3.2-1B-Instruct-bnb-4bit is almost the same speed than a meta-llama/Llama-3.2-3B-Instruct.
Maybe I'm missing something else to make this model work properly

mduran159 changed discussion status to closed
mduran159 changed discussion title from This stuff is so slow, even slower than original model. to I think this model is slower than original model.
Unsloth AI org

What are you using this model on?

shimmyshimmer changed discussion title from I think this model is slower than original model. to Model speed

Sign up or log in to comment