Skip to content

chore: one number for the model size — 2.59GB, not 2.4GB in five comments - #21

Merged
karolkulesza merged 1 commit into
mainfrom
chore/one-model-size-figure
Aug 23, 2026
Merged

karolkulesza merged 1 commit into
mainfrom
chore/one-model-size-figure

Conversation

@karolkulesza

Copy link
Copy Markdown
Owner

Five comments described the shipped Gemma 4 E2B artifact as "2.4GB" while the README and every deep dive say 2.59GB. Both were right — 2.59e9 bytes is 2.41 GiB — but nothing in the repo said so, so a reader comparing pubspec.yaml to the README had two sizes for one file and no way to reconcile them.

Normalized on the decimal figure: it is what litert-community publishes, and what the docs already use fifteen times.

File Was Now
pubspec.yaml a 2.4GB artifact is hashed 2.59GB
lib/services/models/model_downloader.dart a 2.4GB artifact must not 2.59GB
integration_test/model_provisioning_test.dart A 2.4GB artifact over a real link 2.59GB
android/.../backup_rules.xml weights (0.5-2.4GB) (0.58-2.59GB)
android/.../data_extraction_rules.xml 0.5-2.4GB of re-downloadable 0.58-2.59GB

The two Android backup-rule files carried a range rather than a stale single value. It spans the Gemma 3 1B fallback to Gemma 4 E2B, so it is now the two sizes those models are documented at — 0.58GB in docs/model-provisioning.md, 2.59GB in docs/on-device-inference.md — rather than a band rounded off both ends.

git grep -E "2\.4 ?GB" now returns nothing.

Comments only; no executable line changed. dart format --set-exit-if-changed reports 0 changed and flutter analyze no issues. The host suite was not rerun, since no executable line moved.

…ents

Five comments described the shipped artifact as "2.4GB" while the README and
every deep dive say 2.59GB. Both were right — 2.59e9 bytes is 2.41 GiB — but
nothing in the repo said so, so a reader comparing pubspec.yaml to the README
had two sizes for one file and no way to reconcile them.

Normalized on the decimal figure: it is what litert-community publishes, and
what the docs already use fifteen times.

The two Android backup-rule files carried a range rather than a stale single
value. It spans the Gemma 3 1B fallback to Gemma 4 E2B, so it is now the two
sizes those models are documented at — 0.58GB in model-provisioning.md, 2.59GB
in on-device-inference.md — rather than a band rounded off both ends.

Comments only; no executable line changed. `dart format` reports 0 changed and
`flutter analyze` no issues.
@karolkulesza
karolkulesza merged commit 603cc59 into main Aug 23, 2026
1 check passed
@karolkulesza
karolkulesza deleted the chore/one-model-size-figure branch August 28, 2026 20:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant