Uncensored SFT V2: Semantically Deduplicated Instruction Dataset

An instruction dataset for language model fine-tuning, processed to remove semantically similar prompts. The dataset was created by author kaushik-harsh-99 and was last updated on 2026-05-16. It is a version of a prior dataset where near-identical instructions with different wording have been deduplicated.

Use Cases

• Training instruction-following models based on a curated set of prompts.
• Reducing training data redundancy based on semantic deduplication techniques.
• Studying prompt engineering variations based on the examples of rephrased questions.

Strengths

• Dataset underwent semantic deduplication to reduce redundancy of near-identical prompts.
• Builds upon a prior version that already removed exact duplicates and performed extensive cleaning.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update timestamp is in the future (2026-05-16).

Provenance

Source

huggingface

Collection Method

Semantically deduplicated from a prior merged instruction dataset.

Freshness

Last updated 2026-05-16 05:46:28