Doc images
Images here are used by the built documentation. PNG/PDF in this folder are not ignored by git so they can be committed.
Plot regeneration instructions also appear as HTML comments (<!-- DOC_PLOT: ... -->) in the markdown files that reference each figure. MkDocs does not render those comments.
Plot screenshots
| File | Where it's used | How to generate |
|---|---|---|
workflow-diagram.png |
index.md, detailed-workflow.md | Render workflow-diagram.tex |
start_workflow_training_convergence.png |
start.md, evaluation-visualization.md | start_workflow.py |
start_workflow_encoder_analysis_split_test.png |
start.md, evaluation-visualization.md | Same |
start_workflow_decoder_probability_shifts_3SG.png |
start.md | Same; decoder eval for 3SG |
data_splits_encoder_by_target_test.png |
data-splits.md | train_sentiment.py |
suite_similarity_heatmap.png |
trainer-suites.md | train_sentiment_positive_suite.py (--write-docs-images) |
suite_cross_encoding_heatmap.png |
trainer-suites.md, cross-model-comparison.md | Same (--write-docs-images) |
symmetric_suite_topk_overlap.png |
trainer-suites.md | train_race_symmetric_suite.py (--write-docs-images) |
symmetric_suite_cross_encoding.png |
trainer-suites.md | Same |
multi_seed_layerwise_similarity.png |
cross-model-comparison.md | python -m gradiend.examples.train_multi_seed_stability --write-docs-images |
multi_seed_encoder_by_target_seeds.png |
multi-seed.md | Same |
multi_seed_encoder_by_target_combined.png |
multi-seed.md | Same |
multi_seed_encoder_by_target_errorbar.png |
multi-seed.md | Same |
multi_seed_encoder_distributions.png |
multi-seed.md | Same |
multi_seed_encoder_scatter.png |
multi-seed.md | Same |
multi_seed_training_convergence.png |
multi-seed.md | Same |
multi_seed_probability_shifts.png |
multi-seed.md | Same |
seed_comparison_topk_overlap.png |
cross-model-comparison.md | Same |
seed_comparison_decoder_cosine.png |
cross-model-comparison.md | Same |
multi_seed_suite_dispersion_heatmap.png |
trainer-suites.md | Multilingual demo multi-seed suite output, converted from the generated PDF |
topk_overlap_heatmap.png |
evaluation-visualization.md | train_gender_de_detailed.py |
topk_overlap_venn.png |
evaluation-visualization.md | Same |
pruning_analysis_metric_grid.png |
pruning-guide.md | runs/paper/pruning_analysis_metric_grid.pdf |
cross_encoding_example_race_gender_oriented_counterfactual.png |
cross-encoding-matrix.md | Race + English gender subset from the multilingual demo BERT cross-encoding outputs |
cross_encoding_example_race_gender_preanchor.png |
cross-encoding-matrix.md | Same |
cross_encoding_example_race_gender_preanchor_black_black_highlight.png |
cross-encoding-matrix.md | Same |
cross_encoding_example_race_gender_oriented_black_black_highlight.png |
cross-encoding-matrix.md | Same |
cross_encoding_example_race_gender_preanchor_black_asian_highlight.png |
cross-encoding-matrix.md | Same |
cross_encoding_example_race_gender_oriented_black_asian_highlight.png |
cross-encoding-matrix.md | Same |
cross_encoding_example_race_gender_preanchor_asian_asian_highlight.png |
cross-encoding-matrix.md | Same |
cross_encoding_example_race_gender_oriented_asian_asian_highlight.png |
cross-encoding-matrix.md | Same |
cross_encoding_example_race_gender_oriented_white_asian_highlight.png |
cross-encoding-matrix.md | Same |
After adding or updating these files, commit them so deployed docs show the plots.