migrate example configs to canonical attn_implementation
This commit is contained in:
@@ -47,7 +47,7 @@ gradient_checkpointing_kwargs:
|
||||
use_reentrant: false
|
||||
resume_from_checkpoint:
|
||||
logging_steps: 1
|
||||
flash_attention: true
|
||||
attn_implementation: flash_attention_2
|
||||
|
||||
warmup_ratio: 0.1
|
||||
evals_per_epoch:
|
||||
|
||||
@@ -46,7 +46,7 @@ gradient_checkpointing_kwargs:
|
||||
use_reentrant: false
|
||||
resume_from_checkpoint:
|
||||
logging_steps: 1
|
||||
flash_attention: true
|
||||
attn_implementation: flash_attention_2
|
||||
|
||||
warmup_ratio: 0.1
|
||||
evals_per_epoch:
|
||||
|
||||
@@ -44,7 +44,7 @@ gradient_checkpointing: true
|
||||
gradient_checkpointing_kwargs:
|
||||
use_reentrant: true
|
||||
logging_steps: 1
|
||||
flash_attention: true
|
||||
attn_implementation: flash_attention_2
|
||||
|
||||
warmup_ratio: 0.1
|
||||
evals_per_epoch: 1
|
||||
|
||||
Reference in New Issue
Block a user