pfnet/nekomata-7b-pfn-qfin
pfnet/nekomata-7b-pfn-qfin is a 7 billion parameter causal language model developed by Preferred Networks, Inc., fine-tuned for financial text generation. Based on rinna/nekomata-7b, it was continually pre-trained on 370 million tokens of specialized Japanese and English financial datasets. This model excels at generating continuous sentences relevant to finance, offering a context length of 2048 tokens.
Loading preview...
Model Overview
pfnet/nekomata-7b-pfn-qfin is a 7 billion parameter causal decoder-only language model developed by Preferred Networks, Inc. It is a fine-tuned version of the rinna/nekomata-7b base model, specifically optimized for generating continuous sentences in the financial domain. The model was continually pre-trained on 370 million tokens of proprietary Japanese and English financial datasets, ensuring commercial usability.
Key Capabilities & Features
- Financial Domain Specialization: Fine-tuned on extensive financial datasets, making it proficient in generating finance-related text.
- Multilingual Support: Supports both Japanese and English languages.
- Context Length: Processes inputs with a context length of 2048 tokens.
- Commercial Use: Datasets used for fine-tuning were generated by Preferred Networks, ensuring clear commercial usage rights under the Tongyi Qianwen LICENSE AGREEMENT.
Performance
Benchmarking using the Japanese Language Model Financial Evaluation Harness shows competitive performance in financial tasks. The model achieved an overall score of 0.4276, slightly outperforming the base nekomata-7b model's 0.4155 across various financial metrics like chabsa (f1: 0.8127) and security_sales_1 (acc: 0.5088).
Limitations
As with all LLMs, this model may produce inaccurate, biased, or objectionable responses. It is not designed for providing legal, tax, investment, or financial advice. Developers should conduct thorough safety testing and tuning for specific applications.