unsloth icon indicating copy to clipboard operation
unsloth copied to clipboard

Support for Cohere 23 models

Open AlirezaF80 opened this issue 1 year ago • 1 comments

It's a really great multilingual model. Would love being able to train it faster with Unsloth! https://huggingface.co/CohereForAI/aya-23-8B https://huggingface.co/CohereForAI/aya-23-35B

AlirezaF80 avatar May 24 '24 09:05 AlirezaF80

We're working on adding all model support!

shimmyshimmer avatar May 24 '24 10:05 shimmyshimmer

I love to see this rolling. It is much needed for underrepresented languages like Persian.

katebsaber96 avatar May 30 '24 18:05 katebsaber96

Working on all model support!

danielhanchen avatar Jun 01 '24 10:06 danielhanchen

I'm confused, is this possible already ? Since the notebook it's written " # Choose ANY! eg teknium/OpenHermes-2.5-Mistral-7B "

Qualzz avatar Jun 10 '24 18:06 Qualzz

@Qualzz Not yet sorry - but technically Unsloth will check, and will error out if it doesnt worjk

danielhanchen avatar Jun 11 '24 13:06 danielhanchen

Hi guys, apologies for the delays - Cohere models and nearly every model in existence (transformer style) are now supported! :)

Read our blogpost about it: https://unsloth.ai/blog/gemma3#everything

Also FFT + all training methods is now supported

Also multiGPU is coming real soon so be on the lookout!!

We also uploaded the 4bit models and some GGUFs to our hugging face: https://huggingface.co/unsloth

CC: @AlirezaF80 @shimmyshimmer @ewre324 @VertaKhan @akbargherbal @boltonn @oombard @flaviusburca @katebsaber96 @choyakawa @svngoku @Qualzz @avcode-exe

shimmyshimmer avatar Mar 15 '25 10:03 shimmyshimmer