NickIBrody/coffeescript-code-suite

1---2pretty_name: CoffeeScript Code Suite3language:4- code5license:6- mit7- apache-2.08- bsd-2-clause9- bsd-3-clause10- isc11size_categories:12- 10K<n<100K13task_categories:14- text-generation15- translation16tags:17- code18- coffeescript19- javascript20- translation21- completion22- permissive-license23source_datasets:24- original25configs:26- config_name: default27  data_files:28  - split: train29    path: train.jsonl30  - split: validation31    path: validation.jsonl32  - split: test33    path: test.jsonl34---35 36# CoffeeScript Code Suite37 38CoffeeScript Code Suite is a public code dataset built from permissively licensed open-source CoffeeScript repositories. It is designed for three practical uses:39 401. CoffeeScript domain adaptation and continued pretraining through `raw_corpus` examples.412. CoffeeScript completion training through `completion` examples.423. CoffeeScript and JavaScript translation training through `coffee_to_js` and `js_to_coffee` examples.43 44The dataset was assembled automatically from public GitHub repositories and includes both current repository snapshots and selected historical file versions from git history. Each record keeps source provenance so downstream users can audit where the code came from.45 46## Dataset Summary47 48- Name: `CoffeeScript Code Suite`49- Suggested Hub repo name: `brody/coffeescript-code-suite`50- Records: `49,491`51- Training split: `44,535`52- Validation split: `2,481`53- Test split: `2,475`54- Source repositories: `79`55- Raw candidate chunks: `14,682`56- Compilable translation chunks: `10,313`57 58### Record Types59 60- `raw_corpus`: raw CoffeeScript snippets for code-language modeling.61- `completion`: prefix-to-continuation CoffeeScript completion samples.62- `coffee_to_js`: CoffeeScript to JavaScript translation pairs.63- `js_to_coffee`: JavaScript to CoffeeScript translation pairs.64 65## Data Schema66 67Each line is a JSON object. Common fields:68 69```json70{71  "id": "owner/repo:path:chunk:type",72  "type": "coffee_to_js",73  "prompt": "Convert this CoffeeScript to JavaScript...\n```coffeescript\n...\n```",74  "response": "...",75  "source_lang": "CoffeeScript",76  "target_lang": "JavaScript",77  "repo": "owner/repo",78  "path": "src/example.coffee",79  "license": "MIT",80  "commit": "abcdef123456",81  "stars": 1234,82  "source_url": "https://github.com/owner/repo/blob/abcdef123456/src/example.coffee",83  "line_start": 1,84  "line_end": 4285}86```87 88Notes:89 90- `raw_corpus` records have an empty `prompt` and the CoffeeScript snippet in `response`.91- `completion` records use a CoffeeScript prefix in `prompt` and the continuation in `response`.92- Translation records contain executable source/target code pairs when compilation succeeded.93 94## Source Collection Policy95 96The dataset builder uses the following collection policy:97 98- public GitHub repositories only99- permissive licenses only100- current repository snapshots plus selected historical CoffeeScript file versions101- CoffeeScript files only102- vendor, build, generated, docs, demos, and example-heavy paths filtered out103- exact deduplication over normalized chunks104 105Accepted license families in this release:106 107- MIT108- Apache-2.0109- BSD-2-Clause110- BSD-3-Clause111- ISC112 113## Construction Process114 115The dataset was built with an automated pipeline that:116 1171. discovers public CoffeeScript repositories1182. clones repositories locally1193. extracts `.coffee` files from repository heads and git history1204. splits files into training-sized chunks1215. compiles CoffeeScript chunks to JavaScript when possible1226. creates completion and translation samples1237. removes exact duplicates1248. splits records into train, validation, and test sets125 126## Quality Notes127 128Strengths:129 130- source provenance is retained for every record131- dataset contains real-world CoffeeScript from multiple projects rather than synthetic-only samples132- includes both direct code modeling and translation-style supervision133- historical versions increase stylistic and temporal diversity134 135Limitations:136 137- this is an automatically generated dataset, not a hand-curated benchmark138- some projects contribute more samples than others139- `raw_corpus` and `completion` include chunks that may not compile in isolation140- translation pairs only exist for chunks that compiled successfully141- exact deduplication is applied, but near-duplicate code may still remain142 143## Intended Uses144 145Recommended uses:146 147- continued pretraining or domain adaptation for code models with CoffeeScript support148- SFT for CoffeeScript completion149- JavaScript/CoffeeScript translation experiments150- retrieval and evaluation experiments on legacy JavaScript ecosystem code151 152Less suitable uses:153 154- legal or security-sensitive compliance analysis155- benchmark claims about general programming ability156- direct deployment without additional filtering, evaluation, and safety review157 158## Split Sizes159 160```text161train.jsonl       44,535162validation.jsonl   2,481163test.jsonl         2,475164total             49,491165```166 167## Build Summary168 169```json170{171  "repos": 79,172  "files": 10586,173  "chunks": 14682,174  "candidates": 14682,175  "compiled_ok": 10313,176  "compile_failed": 4369,177  "examples": 49491,178  "train": 44535,179  "validation": 2481,180  "test": 2475181}182```183 184## Provenance185 186Each record contains:187 188- repository name189- repository-relative path190- source commit191- source URL192- license label193 194This makes the dataset auditable and easier to extend or prune.195 196## Citation197 198If you use this dataset, cite the dataset repository and the original upstream repositories referenced in the provenance fields.199