Chiboard 1 Sft: Large-Scale Chinese IME Training Data

59,293,523 examples train models to convert pinyin and mixed active buffers into Chinese text within a rolling decode-window IME. The dataset, created by johnbean393, is split into training, test, and development sets. It was last updated on Hugging Face in July 2026.

Use Cases

• Train a pinyin-to-Hanzi conversion model based on the supervised fine-tuning task described.
• Evaluate IME model performance on committed context and active buffer prediction tasks.
• Benchmark rolling decode-window IME systems using the provided train/test/dev splits.
• Research Chinese language model behavior in interactive text entry scenarios.

Strengths

• 59,293,523 total rows provide a substantial volume for training.
• Explicit splits for training (56,918,955 rows), testing (1,184,732 rows), and development (1,189,836 rows) facilitate structured model evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

johnbean393 on Hugging Face

Freshness

Last updated 2026-07-07 19:25:51; freshness should be verified.