Liger-Kernel icon indicating copy to clipboard operation
Liger-Kernel copied to clipboard

[New Model] Support LLaMA 4 family

Open BenasdTW opened this issue 7 months ago • 2 comments

🚀 The feature, motivation and pitch

The Llama 4 models are auto-regressive language models that use a mixture-of-experts (MoE) architecture and incorporate early fusion for native multimodality. https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct

Alternatives

No response

Additional context

No response

BenasdTW avatar Apr 06 '25 11:04 BenasdTW

I will try to do this

vulkomilev avatar Apr 15 '25 20:04 vulkomilev

is there any progress on this?

kir152 avatar May 02 '25 13:05 kir152

Hello, I am working on this. Here is my draft: #740

Manan17 avatar Jun 02 '25 02:06 Manan17

The Liger support for LLama4 family is now added.

Manan17 avatar Jul 08 '25 21:07 Manan17

Thanks!

BenasdTW avatar Jul 09 '25 04:07 BenasdTW