Nanthasit/sakthai-context-0.5b-merged

TEXT GENERATIONConcurrent Unit Cost:1Model Size:0.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 5, 2026Architecture:Transformer Featherless Exclusive Cold

Nanthasit/sakthai-context-0.5b-merged is a 0.5 billion parameter language model based on Qwen2.5-0.5B-Instruct, fine-tuned for enhanced contextual understanding and instruction following. It integrates a LoRA adapter and was trained on the Nanthasit/sakthai-combined-v3 dataset. This model demonstrates strong performance on custom SakThai tasks, including multi-turn recall and tool-calling awareness, making it suitable for applications requiring precise instruction adherence and contextual understanding.

Loading preview...

Model Overview

Nanthasit/sakthai-context-0.5b-merged is a 0.5 billion parameter language model developed by Nanthasit. It is built upon the Qwen2.5-0.5B-Instruct base model and incorporates a LoRA adapter (Nanthasit/sakthai-context-0.5b-tools) for specialized capabilities. The model was fine-tuned using the Nanthasit/sakthai-combined-v3 dataset, which comprises approximately 2153 examples, over 3 epochs.

Key Capabilities and Performance

This model is specifically designed to excel in tasks requiring strong contextual understanding and precise instruction following. It has demonstrated:

  • 100% success rate on custom SakThai evaluation tasks, including multi-turn recall, instruction adherence, JSON output generation, and tool-calling awareness.
  • Improved contextual processing due to its specialized training.

While general benchmarks show moderate performance (e.g., PIQA 68.0%, ARC-Easy 56.0%, HellaSwag 40.0%), its strength lies in its fine-tuned ability to handle complex instructions and maintain context over multiple turns.

Ideal Use Cases

This model is particularly well-suited for applications where:

  • Accurate instruction following is critical, especially in multi-step or multi-turn interactions.
  • Contextual recall over extended conversations is necessary.
  • Structured output generation (like JSON) is required.
  • Tool-calling awareness is beneficial for integrating with external functions or APIs.