Nanthasit/sakthai-context-1.5b-merged-v2

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 30, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Nanthasit/sakthai-context-1.5b-merged-v2 is a 1.5 billion parameter merged full-weight checkpoint of Qwen2.5-1.5B-Instruct, developed by Nanthasit. This model is specifically fine-tuned for structured tool-calling and function-calling, offering stronger reasoning than 0.5B models while remaining suitable for CPU/edge inference. It excels at argument filling and multi-turn tool behavior, requiring 4-8 GB RAM for optimal performance.

Loading preview...

SakThai Context 1.5B Merged V2 Overview

Nanthasit/sakthai-context-1.5b-merged-v2 is a 1.5 billion parameter language model based on Qwen2.5-1.5B-Instruct, specifically designed for structured tool-calling and function-calling. It serves as a mid-weight option, balancing stronger reasoning capabilities compared to 0.5B models with a smaller footprint than 7B models, making it ideal for CPU/edge inference environments.

Key Capabilities & Features

  • Optimized for Tool-Calling: Fine-tuned to produce structured <tool> / <tool_call> output, including JSON arguments.
  • Mid-Weight Performance: Offers enhanced reasoning over 0.5B models while being more resource-efficient than 7B alternatives, requiring approximately 4-8 GB RAM.
  • Deployment Flexibility: Available in BF16 safetensors and GGUF Q4_K_M formats, supporting llama.cpp and Hugging Face Transformers for CPU-only or edge deployments.
  • Context Window: Features a substantial 32,768 token context window.
  • Internal Benchmarks: Achieves 34.9% on SakThai Bench v2 selection and 44.2% on arguments, with 100% multi-tool correctness and valid JSON in internal smoke tests.

When to Use This Model

This model is best suited for applications requiring reliable tool-calling and function-calling on devices with limited resources, such as laptops or desktops. It's a strong choice when you need more robust argument filling and multi-turn tool behavior than a 0.5B model can provide, but a 7B model is too resource-intensive. Note that the model requires a <tools> XML block in the system prompt for tool-calling behavior and is primarily tested for English-only tool calls.

Finetunes
1 models