sabaridsnfuji/Qwen3-1.7B-tamil-16bit-Instruct

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 18, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Warm

The sabaridsnfuji/Qwen3-1.7B-tamil-16bit-Instruct is a 1.7 billion parameter Qwen3-based instruction-tuned language model developed by sabaridsnfuji. This model is specifically fine-tuned for the Tamil language, demonstrating strong fluency and minimal hallucinations. It is optimized for general-purpose tasks in Tamil, offering a solid foundation for Tamil-focused LLM applications.

Loading preview...

Overview

The sabaridsnfuji/Qwen3-1.7B-tamil-16bit-Instruct is a 1.7 billion parameter language model based on the Qwen3 architecture, developed by sabaridsnfuji. It has been specifically fine-tuned for the Tamil language, leveraging Unsloth and Huggingface's TRL library for accelerated training. This model represents an initial step in building a suite of Tamil-focused LLMs, with larger 4B and 8B versions currently under development.

Key Capabilities

  • Tamil Language Proficiency: Achieves 91.4% fluency across diverse tasks in Tamil.
  • High Quality Score: Demonstrates a 70.4% overall quality score.
  • Safety and Reliability: Features a 98.4% safety rating, indicating minimal hallucinations.
  • Efficient Training: Fine-tuned using Unsloth, enabling 2x faster training.

Good For

  • Tamil-centric Applications: Ideal for use cases requiring strong performance in the Tamil language.
  • Instruction Following: Designed to respond effectively to instructions in Tamil.
  • Early-stage Development: Suitable for developers looking to integrate a specialized Tamil LLM into their projects.

Finetunes
1 models