---
format: "aidr-story-markdown/v1"
id: "8429d4e75e515ea32a091c2acf135abb560d2abe3c4fa69e9976d7da4125ad5f"
canonical_url: "https://aidr.today/8429d4e7?lang=en"
title: "FineDNA Releases 300B Token Dataset for LLM Genomic Training"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-01T13:07:08.000Z"
category: "Data"
topics: ["llm","data-engineering"]
source_urls: ["https://x.com/LoubnaBenAllal1/status/2116785229159673914","https://x.com/_lewtun/status/2116817535932789059"]
summary: "FineDNA released a 300B token dataset for LLM genomic training. The training set includes genomic sequences, research papers, and mixed DNA text, along with 200k supervised fine-tuning samples and approximately 100k reinforcement learning samples. Training the Qwen3-14B model on this dataset produces the strongest open model for DNA understanding."
---

# FineDNA Releases 300B Token Dataset for LLM Genomic Training

> [Open the canonical story](<https://aidr.today/8429d4e7?lang=en>)

**Published:** 2026-11-01T13:07:08.000Z
**Category:** Data
**Topics:** llm, data\-engineering

## Summary

FineDNA released a 300B token dataset for LLM genomic training\. The training set includes genomic sequences, research papers, and mixed DNA text, along with 200k supervised fine\-tuning samples and approximately 100k reinforcement learning samples\. Training the Qwen3\-14B model on this dataset produces the strongest open model for DNA understanding\.

## Sources

- [Story source](<https://x.com/LoubnaBenAllal1/status/2116785229159673914>)
- [Supporting source](<https://x.com/_lewtun/status/2116817535932789059>)