fix: wrong parameter for flash attention in pickle __getstate__

This commit is contained in:
Andrei Betlen 2024-04-30 09:32:47 -04:00
parent 22d77eefd2
commit 29b6e9a5c8

View file

@ -1777,7 +1777,7 @@ class Llama:
logits_all=self.context_params.logits_all,
embedding=self.context_params.embeddings,
offload_kqv=self.context_params.offload_kqv,
flash_offload=self.context_params.flash_offload,
flash_attn=self.context_params.flash_attn,
# Sampling Params
last_n_tokens_size=self.last_n_tokens_size,
# LoRA Params