GGearDel.
한국어

Tools

Browser tools for Korean AI data

These pages run in the tab. Pastes and local files are not uploaded. GearDel still does not host dataset archives. Follow the groups in order for a Korean SFT file you already have.

Pick a dataset by task

Filter catalog rows by job and commercial mark. Not a paste box.

Compare two catalog rows

Side-by-side license, provider, and size strings already stored.

Bring data in

Move between tables and JSONL. Hangul folding lives under Inspect.

Korean characters and bytesCount UTF-8 and EUC-KR bytes in the browser. Missing characters are not guessed.Open →

Inspect

Check format, fields, excerpts, citations, Hangul folding, quality, scripts, scrape leftovers, and tokens in this tab.

Hangul normalize and quality

NFC, mojibake, scripts, scrape leftovers. URLs stay as they are.

Instruction JSONL checker and converterCheck and convert Alpaca, ShareGPT, and OpenAI JSONL. Stays in the browser.Open →JSONL field viewer and row samplerView JSONL fields and sample rows in the browser. No upload.Open →Open catalog excerpts, not the filesOpen catalog excerpts only. GearDel still does not host dataset files.Open →Build a citation from catalog fieldsBuild a citation from catalog fields. Authors are not invented.Open →Token count for Korean textCount o200k (GPT-4o, GPT-5) and cl100k. No GPT-6 ranks, so no invented count.Open →

Clean

Mask PII patterns, drop copies, cut long rows.

Mask Korean phone, RRN, and emailMask Korean phone, RRN, and email patterns in the browser. No upload.Open →Drop duplicate JSONL linesDrop duplicate JSONL by prompt or NFC. Stays in the browser.Open →Filter JSONL by token lengthFilter JSONL by cl100k or o200k token length. No upload.Open →

Split and eval

Seed-split, then check overlap with eval prompts.

Split JSONL into train, val, and testSplit JSONL into train, val, and test with a seed hash. Browser only.Open →Whether train JSONL overlaps eval promptsFind train JSONL overlapping eval prompts in the browser. No upload.Open →

Train wrap

Wrap published chat strings, then count tokens.

Wrap a chat template and count tokensWrap chat in ChatML, Llama 3, or Gemma and count tokens. No upload.Open →

Alpaca vs ShareGPT vs messages·Eval vs train