Open LamOne1 opened 1 year ago
This would require adding support for ALiBi. This is not a priority at the moment.
A related feature request is https://github.com/Lightning-AI/lit-gpt/issues/199
+1
Flash attention now supports Alibi btw, so in that case, Alibi support in Lit-GPT could be a bit easier.
We don't support flash attention from flash-attn
. Supporting alibi would warrant an entirely new model_alibi.py
definition
I think we should update this perhaps:
We support flash attention via PyTorch's scaled_dot_product_attention
, just not via Tri Dao's flash-attn
. The former uses one of the latter's implementations internally.
I'd like to request a support for BLOOM as it was pretrained on many languages