pfnet/Qwen3-1.7B-pfn-qfin

Hugging Face
TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 15, 2025License:otherArchitecture:Transformer Featherless Exclusive Warm

Qwen3-1.7B-pfn-qfin is a 1.7 billion parameter causal language model developed by Preferred Networks, fine-tuned from Qwen/Qwen3-1.7B-Base. This model is specifically optimized for financial domain tasks, having been fine-tuned on approximately 400 million tokens from proprietary datasets. It demonstrates improved performance on Japanese financial evaluation benchmarks compared to its base model, making it suitable for applications requiring financial text generation and analysis.

Loading preview...

Overview

Qwen3-1.7B-pfn-qfin is a 1.7 billion parameter language model developed by Preferred Networks, Inc. It is a fine-tuned version of the Qwen3-1.7B-Base model, specifically enhanced for financial applications. The model was trained on approximately 400 million tokens from multiple specialized datasets provided by Preferred Networks, ensuring commercial usability.

Key Capabilities

  • Financial Domain Specialization: Fine-tuned on extensive financial datasets to improve performance in financial contexts.
  • Enhanced Japanese Financial Understanding: Demonstrates superior accuracy and F1 scores on the Japanese Language Model Financial Evaluation Harness across various tasks like chabsa, cma_basics, and fp2, significantly outperforming the base Qwen3-1.7B model.
  • Causal Language Modeling: Designed for generating continuous and coherent text.

Benchmarking Highlights

The model shows notable improvements in financial benchmarks:

  • chabsa (f1): 0.7116 (vs. 0.5734 for Qwen3-1.7B)
  • cma_basics (acc): 0.5263 (vs. 0.3158 for Qwen3-1.7B)
  • OVER ALL: 0.4513 (vs. 0.3527 for Qwen3-1.7B)

Usage Considerations

This model is released under the PLaMo Community License. Developers should conduct their own safety testing and tuning for specific applications, as the model's outputs cannot be fully predicted and may contain inaccuracies or biases. It is not intended for providing legal, tax, investment, or financial advice.