Nanthasit/sakthai-context-7b-merged

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 6, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Nanthasit/sakthai-context-7b-merged is a 7.6 billion parameter language model developed by Nanthasit, based on the Qwen2.5-7B-Instruct architecture. This model is a full-parameter merged checkpoint, fine-tuned with LoRA adapters for enhanced structured tool-calling and instruction following. It excels in tasks requiring precise adherence to instructions and integration with external functions, making it suitable for agentic applications.

Loading preview...

Model Overview

Nanthasit/sakthai-context-7b-merged is the top-performing model within the SakThai Context family, developed by Nanthasit. It is a 7.6 billion parameter model built upon the Qwen/Qwen2.5-7B-Instruct base, utilizing a Qwen2.5 decoder-only transformer architecture. The model was fine-tuned using LoRA adapters, which were subsequently merged into the full checkpoint, specifically targeting improved structured tool-calling and instruction following capabilities.

Key Capabilities

  • Structured Tool-Calling: Supports advanced tool integration using Qwen2.5's tokenizer tool schema, enabling interaction with external functions like get_weather.
  • Instruction Following: Demonstrates strong adherence to user instructions, as evidenced by 100% success in evaluation categories including instruction following and JSON output.
  • Context Recall & Factual Accuracy: Achieved perfect scores in context recall and factual accuracy during internal evaluations.
  • Robust Performance: Evaluated on a Tesla T4, it successfully passed all 8 test categories, including basic response, multi-turn conversations, and name recognition.

Good For

  • Applications requiring reliable tool-calling and agentic behavior.
  • Use cases where precise instruction following and structured output (e.g., JSON) are critical.
  • Developers looking for a 7B-class model with strong general instruction-following capabilities and a focus on integration.