You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am trying to pretrain a MPT model using llm-foundry using AliBi with flash attention. During pre training, I see the below warning -
WARNING: composer.algorithms.alibi.alibi: ALiBi had no effect on the model! Support for ALiBi surgery is currently limited to the following classes:
transformers.models.bert.modeling_bert.BertEmbeddings
transformers.models.bert.modeling_bert.BertSelfAttention
transformers.models.gpt2.modeling_gpt2.GPT2Attention
transformers.models.gpt2.modeling_gpt2.GPT2Model
transformers.models.roberta.modeling_roberta.RobertaEmbeddings
transformers.models.roberta.modeling_roberta.RobertaSelfAttention
I have followed PR#820 for alibi with FA2 for setup, and have used the following in pretrain yaml file -
Just to confirm alibi hasn't been used, I had converted the composer checkpoint to a HF one using scripts/inference/convert_composer_to_hf.py. I find the attn_config.alibi flag is set to False in the config.json file.
Some insights and direction on how to use alibi with flash attention 2 would be immensely helpful.
The text was updated successfully, but these errors were encountered:
I am trying to pretrain a MPT model using llm-foundry using AliBi with flash attention. During pre training, I see the below warning -
I have followed PR#820 for alibi with FA2 for setup, and have used the following in pretrain yaml file -
Just to confirm alibi hasn't been used, I had converted the composer checkpoint to a HF one using
scripts/inference/convert_composer_to_hf.py
. I find theattn_config.alibi
flag is set toFalse
in theconfig.json
file.Some insights and direction on how to use alibi with flash attention 2 would be immensely helpful.
The text was updated successfully, but these errors were encountered: