apunkt/doc-to-lora
1
0
Fork 0
mirror of https://github.com/SakanaAI/doc-to-lora.git synced 2026-07-26 17:11:02 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

scaling gen lora + q gen round for cd + gsm8k

Browse source
This commit is contained in:
51616 2025-09-18 16:02:22 +09:00
parent 2af651fa41
commit 76bf6981ba
7 changed files with 183 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

23
run_eval.py
View file

@ -63,7 +63,7 @@ if __name__ == "__main__":
parser.add_argument(
"--max_test_samples_per_ds",
type=int,
default=1000,
default=500,
help=(
"Maximum number of validation samples per dataset. "
"If -1, uses values from checkpoint config."
@ -102,6 +102,12 @@ if __name__ == "__main__":
action="store_true",
help="Use generated queries for context distillation training.",
)
parser.add_argument(
"--q_gen_rounds",
type=int,
default=4,
help="Number of rounds of query generation for context distillation.",
)
parser.add_argument(
"--use_iterative_mode",
action="store_true",
@ -128,6 +134,21 @@ if __name__ == "__main__":
action="store_true",
help="Add ctx to base model's input",
)
parser.add_argument(
"--truncate_if_too_long_inp",
action="store_true",
help="Truncate input sequences that are too long",
)
parser.add_argument(
"--truncate_if_too_long_ctx",
action="store_true",
help="Truncate ctx sequences that are too long",
)
parser.add_argument(
"--gen_lora_scaling",
type=float,
default=1.0,
)
cli_args = vars(parser.parse_args())
# setup_logging(output_dir, debug=os.getenv("DEBUG", False))