Skip to content

[adding name to NLL evaluator] #104

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

Merged
merged 6 commits into from
Nov 23, 2025
Merged

[adding name to NLL evaluator] #104

Changes from 4 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
61ff912
adding name to nll evaluator
ruiqi-zhong Nov 20, 2025
ec7bc3d
Merge branch 'main' into nll_eval_name
ruiqi-zhong Nov 20, 2025
64dc79b
b
ruiqi-zhong Nov 20, 2025
be8a62c
b
ruiqi-zhong Nov 20, 2025
9fd9678
default eval name to test
ruiqi-zhong Nov 23, 2025
59eb13d
b
ruiqi-zhong Nov 23, 2025
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
10 changes: 6 additions & 4 deletions tinker_cookbook/supervised/nll_evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,8 @@


class NLLEvaluator(TrainingClientEvaluator):
def __init__(self, data: list[tinker.Datum]):
Copy link
Collaborator

@joschu joschu Nov 21, 2025

Choose a reason for hiding this comment

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

actually wouldn't it be better to make the name default to "test", and remove the prefixing of "test" from the training loop?

Copy link
Contributor Author

@ruiqi-zhong ruiqi-zhong Nov 23, 2025

Choose a reason for hiding this comment

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

yeah that's better -- I thought the original code meant to always want "test/" to be the prefix of the log name but that's indeed a bit cumbersome.

def __init__(self, data: list[tinker.Datum], name: str | None = None):
self.name = name
self.data = data

async def __call__(self, training_client: tinker.TrainingClient) -> dict[str, float]:
Expand All @@ -16,9 +17,10 @@ async def __call__(self, training_client: tinker.TrainingClient) -> dict[str, fl
logprobs = [x["logprobs"] for x in result.loss_fn_outputs]
weights = [datum.loss_fn_inputs["weights"] for datum in self.data]
nll = compute_mean_nll(logprobs, weights)
return {"nll": nll}
key = "nll" if self.name is None else f"{self.name}/nll"
return {key: nll}

@classmethod
def from_dataset(cls, dataset: SupervisedDataset) -> "NLLEvaluator":
def from_dataset(cls, dataset: SupervisedDataset, name: str | None = None) -> "NLLEvaluator":
all_data = list(itertools.chain(*[dataset.get_batch(i) for i in range(len(dataset))]))
return cls(all_data)
return cls(all_data, name=name)
Loading