NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·2h ago
AI Research
arXiv CS.AI·Jun 10·5 min read
Instruction Finetuning DeepSeek-R1-8B Model Using LoRA and NEFTune
TL;DR
Financial named-entity recognition (NER) struggles with misclassifications in large language models (LLMs). The DeepSeek-R1-8B model was fine-tuned using Low-Rank Adaptation (LoRA) and Noisy Embedding Fine-Tuning (NEFTune) to enhance its performance.
✦ Why It Matters
Engineers can leverage LoRA and NEFTune to improve NER tasks in specialized fields like finance.
Key Takeaways
- Utilize LoRA for efficient model adaptation in specific domains.
- Incorporate NEFTune to improve model generalization during training.
- Leverage instruction-input-output triples for effective fine-tuning.
- Aim for high micro-F1 scores to benchmark model performance.
How It Works
LoRA introduces lightweight matrices into the model's Transformer layers, allowing for efficient adaptation without extensive retraining. NEFTune enhances the training process by adding uniform noise to the embedding vectors, which helps the model learn more robust representations of financial entities.