Kurtis EON1 SFT: 200,000 Samples for Instruction-Tuning

200,000 text samples aggregated from nine public datasets, including HuggingFaceTB/cosmopedia-v2 (32.15%) and teknium/OpenHermes-2.5 (29.62%). The dataset was created by ethicalabs and last updated on March 16, 2026. It appears to be a curated collection for supervised fine-tuning of language models.

Use Cases

• Supervised fine-tuning of language models based on the described instruction-response pairs.
• Training conversational AI agents based on the psychology and mental health therapy data subsets.
• Benchmarking model performance on diverse instruction types from the multiple source datasets.
• Exploring data synthesis and mixture strategies for AI training based on the provided source distribution.

Strengths

• 200,000 total samples provides a substantial base for model training.
• Aggregates data from nine distinct sources, suggesting diversity in content and style.
• The dataset is stored in an optimized Parquet format, which suggests efficient storage and access.

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 source bias inherent to the aggregated Hugging Face datasets.

Provenance

Source

Aggregated from nine public datasets on Hugging Face.

Collection Method

Likely a curated mixture of existing text datasets.

Time Range

null

Freshness

Last updated 2026-03-16 13:22:51; freshness should be verified.

Geography

null