Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

improve how we setup eval/save strategies and steps #547

Merged
merged 3 commits into from
Sep 13, 2023
Merged

improve how we setup eval/save strategies and steps #547

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
731d080
setup save end eval strategies to be consistent with trainer logic
winglian Sep 10, 2023
4e2a0de
add comments
winglian Sep 10, 2023
7faf032
better eval handling
winglian Sep 13, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 18 additions & 6 deletions src/axolotl/utils/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -556,21 +556,33 @@ def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer, total_num_
"sample_packing_efficiency"
] = cfg.sample_packing_eff_est

if cfg.val_set_size == 0:
if cfg.eval_steps and cfg.evaluation_strategy:
# assume if the user set both, they know what they're doing
training_arguments_kwargs["evaluation_strategy"] = cfg.evaluation_strategy
training_arguments_kwargs["eval_steps"] = cfg.eval_steps
elif cfg.val_set_size == 0:
# no eval set, so don't eval
training_arguments_kwargs["evaluation_strategy"] = "no"
elif cfg.evaluation_strategy and cfg.evaluation_strategy in ["epoch", "no"]:
# if explicitly set for epoch, just set, and eval steps don't matter
training_arguments_kwargs["evaluation_strategy"] = cfg.evaluation_strategy
elif cfg.eval_steps:
# steps isn't used w/ epochs
training_arguments_kwargs["evaluation_strategy"] = "steps"
training_arguments_kwargs["eval_steps"] = cfg.eval_steps
else:
# we have an eval set, but no steps defined, use epoch
# we have an eval set, but no steps defined, default to use epoch
training_arguments_kwargs["evaluation_strategy"] = "epoch"

if cfg.save_strategy:
if cfg.save_steps:
# save_steps implies save_strategy of steps
training_arguments_kwargs["save_strategy"] = "steps"
training_arguments_kwargs["save_steps"] = cfg.save_steps
elif cfg.save_strategy:
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

One side effect would be that: individuals who depend on save_strategy: epoch/no and forget to remove save_step would get a different result. However, this change is good as it's more logical.

training_arguments_kwargs["save_strategy"] = cfg.save_strategy
else:
training_arguments_kwargs["save_strategy"] = (
"steps" if cfg.save_steps else "epoch"
)
# default to saving each epoch if not defined
training_arguments_kwargs["save_strategy"] = "epoch"

if cfg.do_bench_eval:
training_arguments_kwargs["do_bench_eval"] = cfg.do_bench_eval
Expand Down