Bashqort Alpaca: Bashkir Instruction-Response Pairs for LLM Tuning

52,000 instruction-response pairs were translated from Russian to Bashkir for adapting large language models. The dataset was created by author metuKKhud and a small subset of about 10 examples was manually validated by a native speaker. It was last updated on May 20, 2026.

Use Cases

• Instruction tuning of large language models based on the translated instruction-response pairs.
• Evaluating machine translation quality for the Russian-to-Bashkir language pair.
• Training or benchmarking models for Bashkir language understanding and generation tasks.

Strengths

• Derived from a substantial base of 52,000 instruction-response pairs.
• Includes a small subset (~10 examples) that was manually validated by a native Bashkir speaker.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect translation bias inherent to the machine translation process from Russian.

Provenance

Source

Alpaca dataset, specifically the Russian Alpaca (cleaned) from dOrj/alpaca-cleaned-ru.

Collection Method

Translated from Russian to Bashkir.

Freshness

Last updated 2026-05-20 13:59:06; freshness should be verified.