mirror of
https://github.com/hiyouga/LLaMA-Factory.git
synced 2025-12-29 10:10:35 +08:00
[optim] add support to APOLLO (#6617)
Former-commit-id: 5a252e5a458457adbd19da3b68a3897ad2962824
This commit is contained in:
@@ -147,6 +147,7 @@ class Runner:
|
||||
shift_attn=get("train.shift_attn"),
|
||||
report_to="all" if get("train.report_to") else "none",
|
||||
use_galore=get("train.use_galore"),
|
||||
use_apollo=get("train.use_apollo"),
|
||||
use_badam=get("train.use_badam"),
|
||||
use_swanlab=get("train.use_swanlab"),
|
||||
output_dir=get_save_dir(model_name, finetuning_type, get("train.output_dir")),
|
||||
@@ -223,6 +224,13 @@ class Runner:
|
||||
args["galore_update_interval"] = get("train.galore_update_interval")
|
||||
args["galore_scale"] = get("train.galore_scale")
|
||||
args["galore_target"] = get("train.galore_target")
|
||||
|
||||
# apollo config
|
||||
if args["use_apollo"]:
|
||||
args["apollo_rank"] = get("train.apollo_rank")
|
||||
args["apollo_update_interval"] = get("train.apollo_update_interval")
|
||||
args["apollo_scale"] = get("train.apollo_scale")
|
||||
args["apollo_target"] = get("train.apollo_target")
|
||||
|
||||
# badam config
|
||||
if args["use_badam"]:
|
||||
|
||||
Reference in New Issue
Block a user