Labelled, synthetic datasets in the shapes ML pipelines expect (JSONL for text tasks, image annotations, embeddings, and sample weights) for testing data loaders, tokenizers, and training tooling.
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 labelled sentiment-classification dataset in JSON Lines: 24 short product-review-style sentences balanced across positive, negative, and neutral. Fully synthetic; a fixture for testing text-classification loaders, tokenizers, and JSONL parsers.
A token-classification dataset in JSON Lines: 16 tokenized sentences with aligned BIO tags for person, organisation, and location entities. All names, companies, and places are fictional. A fixture for NER model training and sequence-labelling tooling.
A chat fine-tuning dataset in the OpenAI JSONL format: one conversation per line as a messages array with system, user, and assistant turns. Synthetic Q&A content. Paired with an Anthropic-format twin for testing format converters.
The same synthetic conversations in the Anthropic Messages JSONL shape: a top-level system prompt plus a messages array of user and assistant turns. The format twin of the OpenAI file, for testing chat-format conversion.
An instruction-tuning dataset in the Alpaca JSONL format: 20 instruction / input / output triples covering small transformations, extraction, and factual answers. Fully synthetic; a fixture for supervised fine-tuning pipelines.
An extractive question-answering dataset in the SQuAD v2.0 JSON structure, titled articles with context paragraphs, questions, character-offset answers, and one deliberately unanswerable question. Synthetic content; a fixture for QA model training and SQuAD-format loaders.
An abstractive-summarization dataset in JSON Lines: 15 short synthetic news-style documents each paired with a one-sentence summary. A fixture for training and evaluating summarization models and for testing JSONL ingestion.
An English↔Spanish parallel corpus in JSON Lines: 20 aligned sentence pairs of everyday phrases. A fixture for training and evaluating machine-translation models and for testing UTF-8 handling of accented characters.
A set of 24 L2-normalised 16-dimensional text embeddings as JSON: each record pairs an id and its source text with a float vector. A fixture for testing vector stores, similarity search, and embedding loaders. Parquet and .npy twins included.
The same 16-dimensional embeddings as Apache Parquet: id and text columns plus one column per dimension. The columnar twin, for testing analytics engines and Parquet-based vector pipelines.
The embeddings as a raw NumPy array: a 24×16 float32 matrix in .npy format, loadable with numpy.load. The binary twin of the JSON and Parquet files, for testing tensor and matrix loaders.
A simple rendered street scene with a person, a car, and a tree at known pixel coordinates: the image the COCO, YOLO, and Pascal-VOC annotation twins describe. A fixture for testing object-detection loaders and annotation converters.
The class-name list for the detection scene, one label per line: index equals the zero-based line number, matching the YOLO class ids. A companion to the COCO/YOLO/VOC annotation files.
Object-detection annotations for the scene in the COCO JSON format: images, categories, and per-object bounding boxes as [x, y, width, height]. Grouped with YOLO and Pascal-VOC twins for testing annotation-format conversion.
The same detection boxes in the YOLO text format: one object per line as class id and box centre, width, and height normalised to 0–1. The format twin of the COCO and VOC annotations.
The same detection boxes in the Pascal VOC XML format: a per-image annotation with size, and one object element per box with pixel corner coordinates. The XML twin of the COCO and YOLO annotations.
A genuinely-valid safetensors file with two small float32 tensors (36 parameters total): an 8×4 weight and a length-4 bias. The values are meaningless sample data, not a trained model; a fixture for testing safetensors loaders and weight inspectors.
JSON Lines ASR training/eval set for the Wave B synthetic digit utterances: each row points at a clean WAV and carries the expected transcript.
Six synthetic image-caption pairs in JSON Lines, for testing captioning loaders and vision-language eval harnesses.
Short VQA question/answer pairs referencing library images: a fixture for VQA loaders and eval scripts.
A short Markdown document in the mini RAG corpus (doc-a.md). Retrieve against the paired queries JSON.
A short Markdown document in the mini RAG corpus (doc-b.md). Retrieve against the paired queries JSON.
A short Markdown document in the mini RAG corpus (doc-c.md). Retrieve against the paired queries JSON.
A short Markdown document in the mini RAG corpus (doc-d.md). Retrieve against the paired queries JSON.
Retrieval-eval queries with relevant_docs pointers into the mini RAG Markdown corpus.
A synthetic RGB scene with three coloured shapes: input for semantic-segmentation models. Pair with the indexed mask twin.
Indexed 8-bit mask (0=background, 1–3=shapes) for the semantic-segmentation scene twin.
Class-id to name map for the semantic-segmentation mask (background + three shapes).
Named-entity spans over fictional SAMPLE PII sentences, for redaction/NER tooling tests.
Synthetic warehouse detection scene with 3 labelled objects at known coordinates.
COCO JSON with bounding boxes and polygon segmentation for the warehouse scene.
YOLO-format normalised boxes for the warehouse detection scene.
Synthetic retail detection scene with 3 labelled objects at known coordinates.
COCO JSON with bounding boxes and polygon segmentation for the retail scene.
YOLO-format normalised boxes for the retail detection scene.
Synthetic aerial detection scene with 2 labelled objects at known coordinates.
COCO JSON with bounding boxes and polygon segmentation for the aerial scene.
Combined class-name list for Wave F warehouse, retail, and aerial detection scenes.
Sample keypoint annotations (visible/occluded flags) for pose-estimation loader tests.
Multi-label classification JSONL with two labels per row.
Hierarchical category labels for taxonomy-aware classifiers.
Imbalanced label distribution: 2 rare vs 8 common rows.
Zero-shot candidate label lists per text for NLI-style classifiers.
Intent classification JSONL for dialog systems.
Topic classification rows for news routing tests.
Language identification JSONL with EN/ES pairs.
Binary toxicity labels for moderation model eval.
Query-document relevance grades for learning-to-rank training and eval.
Pairwise doc preferences for reward-model and reranker training tests.
Listwise ranked doc lists for nDCG and listwise loss tests.
OpenAI-style tool-call JSONL with function invocation and tool result messages.
Anthropic-style tool_use blocks with paired tool results for agent harness tests.
Multi-step agent action/observation trace for ReAct-style eval harnesses.
L2-normalised 8-dimensional text embeddings in JSON, for vector store loader tests.
NumPy matrix twin of the Wave F 8-dim embeddings.
ID list and model metadata for the Wave F embedding set.
EN/ES sentence embeddings for cross-lingual similarity tests.
int8-quantised embedding vector for quantised vector search tests.
EN image captions referencing Wave F detection scenes.
ES image captions referencing Wave F detection scenes.
FR image captions referencing Wave F detection scenes.
DE image captions referencing Wave F detection scenes.
JA image captions referencing Wave F detection scenes.
Parallel EN/ES/FR/DE captions per image for multilingual eval.
Pascal VOC XML annotations for the warehouse detection scene.
Label id/name map companion for Wave F classification JSONL fixtures.
Tiny SAMPLE prompt fixture (system-concise) for harness and eval wiring.
Tiny SAMPLE prompt fixture (system-json-only) for harness and eval wiring.
Tiny SAMPLE prompt fixture (user-summarize) for harness and eval wiring.
Tiny SAMPLE prompt fixture (user-classify) for harness and eval wiring.
Tiny SAMPLE prompt fixture (user-extract-entities) for harness and eval wiring.
Tiny SAMPLE prompt fixture (fewshot-translate) for harness and eval wiring.
