RealFeedApp The mid-training dataset volume is approximately 950 GB
FineDNA data was used across all three training stages of Carbon-X, including SFT and RL. The mid-training dataset volume is approximately 950 GB.
- Field
- training language models
- What they did
- Created the FineDNA dataset, which was used to train the Carbon-X model (including Qwen3-14B) to read DNA sequences across all training stages.
- Why it matters
- The data consists of approximately 950 GB of configurations for mid-training and mixtures for supervised fine-tuning and reinforcement learning stages.
#llm#dna#carbon-x#qwen#finedna
