pfnet/Qwen2.5-1.5B-pfn-qfin

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:1.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jul 14, 2025License:plamo-community-licenseArchitecture:Transformer Warm

pfnet/Qwen2.5-1.5B-pfn-qfin is a 1.5 billion parameter causal language model developed by Preferred Networks, fine-tuned from Qwen/Qwen2.5-1.5B. This model is specifically optimized for Japanese financial evaluation tasks, demonstrating improved performance on benchmarks like chabsa and cpa_audit. It is designed for generating continuous sentences within a 32768 token context length, leveraging specialized datasets for commercial use.

Loading preview...

Model Overview

pfnet/Qwen2.5-1.5B-pfn-qfin is a 1.5 billion parameter language model developed by Preferred Networks, fine-tuned from the Qwen2.5-1.5B base model. It has been trained on approximately 400 million tokens from proprietary datasets, focusing on Japanese financial contexts, and is released under the PLaMo Community License. The fine-tuning was conducted with a 2048 context length, making it suitable for tasks requiring moderate context understanding.

Key Capabilities & Performance

This model excels in generating continuous sentences and shows enhanced performance on specific Japanese financial evaluation benchmarks. Benchmarking against the base Qwen2.5-1.5B model using the Japanese Language Model Financial Evaluation Harness demonstrates its specialized improvements:

  • chabsa (F1 score): Improved from 0.7269 to 0.7578
  • cma_basics (accuracy): Improved from 0.3684 to 0.3947
  • cpa_audit (accuracy): Improved from 0.1382 to 0.2111
  • fp2 (accuracy): Improved from 0.4035 to 0.4386

Overall, the model achieved an average accuracy of 0.4089 across these financial tasks, compared to 0.3767 for the base model.

Good For

  • Japanese Financial Text Generation: Ideal for applications requiring text generation or analysis within Japanese financial domains.
  • Commercial Use: The model is fine-tuned on commercially clear datasets, making it suitable for business applications under the PLaMo Community License.
  • Sentence Completion: Its base model characteristic of generating continuous sentences is maintained and enhanced for specialized content.

Limitations

As with all LLMs, this model may produce inaccurate, biased, or objectionable responses. It is not designed for providing legal, tax, investment, financial, or other professional advice. Developers should conduct thorough safety testing and tuning for specific applications.

Finetunes
1 models