SwayAlgo/SwayAlgo-Finance-gemma-4-E4B-it-1

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 29, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

SwayAlgo Finance Gemma 4 E4B-it 1 is a 7.9 billion parameter finance-oriented language model developed by SwayAlgo, adapted from google/gemma-4-E4B-it. It is specifically fine-tuned for practical reasoning across various Indian finance domains, including quantitative aptitude, accounting, and banking. The model achieves 45.45% accuracy on the BhashaBench-Finance evaluation, demonstrating strong performance in finance-specific multiple-choice tasks.

Loading preview...

SwayAlgo Finance Gemma 4 E4B-it 1: Finance-Oriented Language Model

SwayAlgo Finance Gemma 4 E4B-it 1 is a specialized language model developed by SwayAlgo, built upon the google/gemma-4-E4B-it base. This 7.9 billion parameter model is meticulously fine-tuned to excel in Indian finance-related reasoning tasks, addressing areas such as quantitative aptitude, accounting, banking, commerce, and regulatory questions.

Key Capabilities and Features

  • Finance Domain Expertise: Optimized for practical reasoning across diverse Indian finance topics, including policy, education, and business operations.
  • Strong Benchmark Performance: Achieves 45.45% accuracy on the full BhashaBench-Finance evaluation, significantly outperforming its base model (39.98%) and other comparable models like FinanceParam (31.42%).
  • Targeted Training: Fine-tuned using a compact LoRA adapter and a focused finance-reasoning corpus, with a strategy to improve multiple-choice performance and numerical consistency.
  • Reproducible Evaluation: Includes full merged model weights, tokenizer, chat-template files, and detailed evaluation artifacts (summary, per-question predictions, mistake file) for transparency.
  • Architecture: Inherits the Gemma4ForConditionalGeneration architecture from its base, with 42 hidden layers and a vocabulary size of 262144.

When to Use This Model

This model is ideal for applications requiring high accuracy in finance-specific multiple-choice questions, particularly those related to the Indian financial context. It is designed for developers and researchers focused on building intelligent assistants or tools that need to understand and reason within complex financial scenarios. Its strong performance across various finance domains and difficulty levels makes it a robust choice for specialized financial NLP tasks.

Finetunes
2 models