mirror of https://github.com/InternLM/InternLM
fix(initialize/launch.py): set default value for use_flash_attn (#158)
* add default for use_flash_attn * fix lintpull/160/head
parent
5ee651c2f1
commit
307c4741d1
|
@ -173,6 +173,10 @@ def args_sanity_check():
|
|||
logger.info("+" * 15 + " beta2_scheduler Info " + "+" * 15) # pylint: disable=W1201
|
||||
logger.info(f"beta2_scheduler: {gpc.config.beta2_scheduler}")
|
||||
|
||||
# process the model config
|
||||
if "use_flash_attn" not in gpc.config.model:
|
||||
gpc.config.model._add_item("use_flash_attn", True)
|
||||
|
||||
|
||||
def launch(
|
||||
config: Union[str, Path, Config, Dict],
|
||||
|
|
Loading…
Reference in New Issue