Lm 100M En Sft Data: 100M-Parameter English Language Model Instruction Tuning Dataset

A 100M English language model instruction tuning dataset used for supervised fine-tuning. The dataset, created by Aeryx-ai, combines the shared ChatML instruct dataset, SmolTalk core, and Dolly-15k. It was used in an experiment comparing two ~100M-parameter models with identical architecture and SFT but different pretraining token budgets.

Use Cases

• Supervised fine-tuning of small language models based on the described instruction data.
• Studying the impact of pretraining token budget on model performance based on the described A/B experiment.
• Replicating or analyzing the instruction-tuning process for the described ~100M-parameter models.
• Creating instruction-response pairs for model training based on the ChatML and Dolly-15k sources mentioned.

Strengths

• Dataset was used for a controlled experiment comparing two ~100M-parameter models, suggesting a defined purpose.
• Combines multiple established instruction sources: ChatML, SmolTalk core, and Dolly-15k.
• Last updated on 2026-06-18 17:35:34, indicating recent activity.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.

Provenance

Source

huggingface, author Aeryx-ai

Collection Method

Likely aggregated and processed from the ChatML instruct dataset, SmolTalk core, and Dolly-15k.

Freshness

Last updated 2026-06-18 17:35:34; freshness should be verified.