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

Extract metric_key_prefix during NotebookProgressCallback.on_evaluate #11347

Merged
merged 7 commits into from
Apr 21, 2021
Merged

Extract metric_key_prefix during NotebookProgressCallback.on_evaluate #11347

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
7766bfe
Pass metric_key_prefix as kwarg to on_evaluate
lewtun Apr 20, 2021
9208eff
Replace eval_loss with metric_key_prefix_loss
lewtun Apr 20, 2021
9bb6e0d
Default to "eval" if metric_key_prefix not in kwargs
lewtun Apr 20, 2021
8d4c85e
Add kwargs to CallbackHandler.on_evaluate signature
lewtun Apr 20, 2021
771789e
Revert "Add kwargs to CallbackHandler.on_evaluate signature"
lewtun Apr 20, 2021
58cf37d
Revert "Pass metric_key_prefix as kwarg to on_evaluate"
lewtun Apr 20, 2021
47e5045
Extract metric_key_prefix from metrics
lewtun Apr 20, 2021
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
14 changes: 9 additions & 5 deletions src/transformers/utils/notebook.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@
# limitations under the License.

import collections
import re
import time
from typing import Optional

Expand Down Expand Up @@ -308,7 +309,7 @@ def on_log(self, args, state, control, logs=None, **kwargs):

def on_evaluate(self, args, state, control, metrics=None, **kwargs):
if self.training_tracker is not None:
values = {"Training Loss": "No log"}
values = {"Training Loss": "No log", "Validation Loss": "No log"}
for log in reversed(state.log_history):
if "loss" in log:
values["Training Loss"] = log["loss"]
Expand All @@ -318,13 +319,16 @@ def on_evaluate(self, args, state, control, metrics=None, **kwargs):
values["Epoch"] = int(state.epoch)
else:
values["Step"] = state.global_step
values["Validation Loss"] = metrics["eval_loss"]
Copy link
Member Author

Choose a reason for hiding this comment

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

this line seemed to be redundant since we later loop over metrics.items(), so i removed it

metric_key_prefix = "eval"
for k in metrics:
if k.endswith("_loss"):
metric_key_prefix = re.sub(r"\_loss$", "", k)
_ = metrics.pop("total_flos", None)
_ = metrics.pop("epoch", None)
_ = metrics.pop("eval_runtime", None)
_ = metrics.pop("eval_samples_per_second", None)
_ = metrics.pop(f"{metric_key_prefix}_runtime", None)
_ = metrics.pop(f"{metric_key_prefix}_samples_per_second", None)
for k, v in metrics.items():
if k == "eval_loss":
if k == f"{metric_key_prefix}_loss":
values["Validation Loss"] = v
else:
splits = k.split("_")
Expand Down