iapp/chinda-qwen3-4b

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:May 28, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

The iapp/chinda-qwen3-4b is a 4.02 billion parameter Thai language model developed by iApp Technology, built on the Qwen3-4B architecture. It features advanced thinking capabilities with a unique thinking/non-thinking mode, achieving 98.4% accuracy in Thai language output and 37% better overall performance than alternatives in its category. This model is optimized for sovereign AI applications, mobile/laptop deployment, mathematical calculations, and code assistance, offering fast inference and resource efficiency.

Loading preview...

OpenThai Chinda 4B: A Sovereign Thai AI Model

OpenThai Chinda 4B, developed by iApp Technology, is a 4.02 billion parameter Thai language model based on the cutting-edge Qwen3-4B architecture. It is designed to provide advanced thinking capabilities within the Thai AI ecosystem, emphasizing high accuracy in Thai language output and efficient performance.

Key Features & Capabilities

  • Advanced Thinking Mode: Offers seamless switching between a detailed reasoning mode for complex problems (math, coding) and an efficient non-thinking mode for general dialogue. This unique feature allows for superior reasoning while maintaining efficiency.
  • Exceptional Thai Language Accuracy: Achieves 98.4% accuracy in Thai output, specifically fine-tuned to prevent unwanted foreign language generation and optimized for Thai linguistic patterns.
  • Superior Performance: Benchmarks show OpenThai Chinda 4B achieving 37% better overall performance compared to its nearest 4B parameter alternative, particularly excelling in mathematical reasoning (MATH500) and code generation (LiveCodeBench) for both English and Thai.
  • Open-Source & Commercial Use: Released under the Apache 2.0 License, permitting free commercial use, modification, and distribution.
  • Context Length: Supports a native context length of 32,768 tokens, extendable up to 131,072 tokens with YaRN scaling.

Suitable Use Cases

  • RAG Applications: Ideal for building Retrieval-Augmented Generation systems that require data processing within Thai sovereignty.
  • Edge Computing: Optimized as a Small Language Model for mobile and laptop applications due to its resource efficiency and fast inference.
  • Mathematical & Code Assistance: Demonstrates strong capabilities in mathematical problem-solving and code generation.
  • Resource-Efficient Deployments: Provides very fast inference with minimal GPU memory consumption, suitable for production environments.

Finetunes
2 models