DeepSeek-R1-8B fine-tuned for financial NER with LoRA and NEFTune

Researchers have fine-tuned the DeepSeek-R1-8B language model for financial named-entity recognition (NER) tasks. By employing Low-Rank Adaptation (LoRA) and Noisy Embedding Fine-Tuning (NEFTune), the adapted model achieved a micro-F1 score of 0.912. This performance surpassed several other baseline models, including Llama3-8B and Qwen3-8B, demonstrating the effectiveness of these techniques for domain-specific NER. AI

IMPACT Enhances financial NER capabilities, potentially improving structured data extraction from financial documents.

RANK_REASON This is a research paper detailing the fine-tuning of an existing open-source model with specific techniques for a particular task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Baichuan2-7B
• DeepSeek-R1-8B
• Llama3-8B
• LoRA
• NEFTune
• Qwen3-8B

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →