Instruction Finetuning DeepSeek-R1-8B Model Using LoRA and NEFTune

The paper presents the DeepSeek-R1-8B model, enhanced for financial named-entity recognition (NER) through instruction fine-tuning using Low-Rank Adaptation (LoRA) and Noisy Embedding Fine-Tuning (NEFTune). The model achieves a micro-F1 score of 0.901, which improves to 0.912 with NEFTune, outperforming several baseline models including Llama3-8B and BERT-Base. This advancement is significant for practitioners focused on domain-specific NER, as it demonstrates effective techniques for adapting LLMs to specialized tasks in finance.