Sinhala Translation Dataset: 109,649 English-Sinhala Pairs for Llama-3 Instruction Tuning

SAWithanage created a dataset of 109,649 English to Sinhala translation pairs formatted for instruction tuning. The data is specifically engineered for training Large Language Models like ShenalMax/SinLlama-Llama-3-8B-Merged. It was last updated on May 18, 2026.

Use Cases

• Fine-tuning LLMs for English-to-Sinhala translation based on the described instruction-tuning format.
• Training translation models using the native Llama-3 conversational Chat Template framework mentioned in the description.
• Benchmarking translation accuracy for Sinhala language models using the curated translation pairs.
• Developing multilingual NLP applications that require high-quality Sinhala translation capabilities.

Strengths

• Contains 109,649 translation pairs, providing a substantial corpus for model training.
• Described as 'highly curated, shuffled, and pre-formatted' specifically for instruction tuning.
• Wrapped into the native Llama-3 conversational Chat Template framework for direct compatibility.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but other metadata like file formats, license, and sample data are unavailable.
• Last updated 2026-05-18 02:59:11; freshness should be verified.

Provenance

Source

SAWithanage on Hugging Face.

Collection Method

Likely curated from existing translation resources, but the specific gathering method is not detailed.

Freshness

Last updated 2026-05-18 02:59:11.