Benchmark results, confusion matrices, ROC curves, and classification reports in CSV and JSON, for testing eval dashboards, metric parsers, and leaderboard importers.
A customer-survey response set: 60 respondents rating five statements on a 1–5 Likert scale plus a 0–10 NPS score and a segment. A fixture for testing survey analysis, aggregation, and NPS/CSAT calculations.
A MovieLens-style ratings log: 240 user/item/rating events with Unix timestamps across 50 users and 40 items. A fixture for testing recommender pipelines, collaborative-filtering loaders, and sparse-matrix builders.
A model-evaluation summary in JSON, per-task scores for a fictional model across sentiment, NER, summarization, translation, and QA, each with its metric and sample size. A fixture for testing eval dashboards and leaderboard importers.
A 3-class confusion matrix as CSV: rows are the true class, columns the predicted class, cells the counts. Paired with a JSON twin for testing metric parsers and evaluation visualisers.
The same 3-class confusion matrix as JSON: a labels array plus a nested counts matrix. The structured twin of the CSV, for testing evaluation tooling.
An ROC curve as CSV: decision threshold with the corresponding false-positive and true-positive rates, monotonic from (0,0) to (1,1). A fixture for testing chart tools and AUC calculators.
A per-class classification report in the scikit-learn structure: precision, recall, F1, and support for each class plus accuracy and macro/weighted averages. A fixture for testing metric parsers and report renderers.
Synthetic mAP evaluation summary for object-detection benchmark harness tests.
BM25 score snapshot for comparing neural rerankers against a lexical baseline.
Mean reciprocal rank ground truth for search eval harness tests.
Precomputed DCG/nDCG values for ranking metric unit tests.
Precomputed cosine similarity pairs for embedding eval scripts.
3×3 confusion matrix CSV for classification metric calculators.
JSON twin of the 3-class confusion matrix.
5×5 confusion matrix with off-diagonal noise for multi-class eval.
JSON twin of the 5-class confusion matrix.
Per-class precision/recall/F1 snapshot for classification dashboards.
sklearn-style classification report text for parser snapshot tests.
ROC curve coordinate list for AUC calculator tests.
PR curve coordinates for average-precision metric tests.
Multiple reference captions per image for BLEU/CIDEr caption eval.
Short query list for ranking benchmark harness smoke tests.
Tiny 2×2 cosine similarity matrix for embedding eval unit tests.
Minimal SAMPLE eval metric JSON (accuracy) for dashboard parsers.
Minimal SAMPLE eval metric JSON (f1) for dashboard parsers.
Minimal SAMPLE eval metric JSON (latency_ms) for dashboard parsers.
Minimal SAMPLE eval metric JSON (toxicity) for dashboard parsers.
Minimal SAMPLE eval metric JSON (bleu) for dashboard parsers.
A one-node identity graph with a fixed 1x4 float tensor, useful as the smallest loader and inference smoke test. The matching JSON input and expected output make the model immediately usable in a CI inference assertion.
The named float32 input tensor payload for the Static Identity model, with explicit shapes and JSON values for runner-independent test setup.
The reference-evaluator output for the Static Identity model and supplied input, including shape, dtype, exact values and a 1e-6 tolerance.
A two-node affine transform with fixed weights and bias, exercising initializers, MatMul dimensions and Add broadcasting. The matching JSON input and expected output make the model immediately usable in a CI inference assertion.
The named float32 input tensor payload for the Affine MatMul plus Add model, with explicit shapes and JSON values for runner-independent test setup.
The reference-evaluator output for the Affine MatMul plus Add model and supplied input, including shape, dtype, exact values and a 1e-6 tolerance.
A two-input Add graph where a length-two bias broadcasts across a 2x2 matrix, checking input binding and broadcasting rules. The matching JSON input and expected output make the model immediately usable in a CI inference assertion.
The named float32 input tensor payload for the Two-Input Broadcast Add model, with explicit shapes and JSON values for runner-independent test setup.
The reference-evaluator output for the Two-Input Broadcast Add model and supplied input, including shape, dtype, exact values and a 1e-6 tolerance.
A ReduceSum graph with a symbolic batch dimension and fixed feature width, for dynamic-shape import and batch-output tests. The matching JSON input and expected output make the model immediately usable in a CI inference assertion.
The named float32 input tensor payload for the Dynamic-Batch ReduceSum model, with explicit shapes and JSON values for runner-independent test setup.
The reference-evaluator output for the Dynamic-Batch ReduceSum model and supplied input, including shape, dtype, exact values and a 1e-6 tolerance.
