Skip to content

docs(models): remove misleading Index column from output tables (#1009)#1358

Open
mvanhorn wants to merge 1 commit intogoogle:mainfrom
mvanhorn:docs/1009-remove-model-index-column
Open

docs(models): remove misleading Index column from output tables (#1009)#1358
mvanhorn wants to merge 1 commit intogoogle:mainfrom
mvanhorn:docs/1009-remove-model-index-column

Conversation

@mvanhorn
Copy link
Copy Markdown

Summary

Removes the misleading Index column from the "List of possible outputs" table in every model README. Keeps the second table ("List of possible model's outputs") untouched because those indices are real softmax positions.

Why this matters

From #1009 (opened by @reyammer):

In the models' README, we have a table with the list of possible outputs and IDs. These IDs are NOT the IDs / indexes of the model's softmax layer... We should probably clarify this nit, as it may lead to false assumptions. Update: we should likely just remove them.

@ia0 in thread: "+1 to remove them, they seem to have to significance"

The first table documents the tool-level outputs (MagikaResult.prediction.output.label). Its row numbers were purely positional and were consistently being misread as model-internal IDs. Removing the column makes the contract match reality.

Changes

Removes the Index column from the first table in:

  • assets/models/standard_v2_0/README.md
  • assets/models/standard_v2_1/README.md
  • assets/models/standard_v3_0/README.md
  • assets/models/standard_v3_1/README.md
  • assets/models/standard_v3_2/README.md
  • assets/models/standard_v3_3/README.md
  • python/src/magika/models/standard_v3_3/README.md (mirrored copy)

The second table in each v3_x README ("List of possible model's outputs") is left as-is because those indices legitimately correspond to model outputs per the existing prose.

standard_v1 has no such table and is unchanged.

Testing

Diff is a pure column drop. No generator script produces these files (grepped python/scripts/sync.py, assets/, and scripts directories for any "List of possible outputs" producer, none found) so no generator update is needed.

Fixes #1009

This contribution was developed with AI assistance (Codex + Claude Code).

…le#1009)

The Index column in the 'List of possible outputs' table implied these
numbers were the model's softmax indexes, but they are not. The real
source of truth is the model's config target labels.

Removes the Index column from the first table in each model README
(standard_v2_0, v2_1, v3_0, v3_1, v3_2, v3_3 and the mirrored copy
under python/src/magika/models/). The second table ('List of possible
model's outputs') intentionally uses real model indices and is
unchanged.

Per maintainer consensus in google#1009:
  @reyammer: 'we should likely just remove them'
  @ia0: '+1 to remove them'

Fixes google#1009
@mvanhorn mvanhorn requested a review from reyammer as a code owner April 17, 2026 12:31
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.

Clarify in models README's that the IDs in the "model's outputs" are not the real IDs (or, better, remove them)

1 participant