JibayAi/Jibay_2
JibayAi/Jibay_2 is a 2 billion parameter, open-source language model built upon the Qwen 3 family, featuring a 32,768 token context length. Designed to be simple, trainable, and customizable, it serves as an efficient alternative for developers needing a controllable model without excessive computational overhead. This model excels at fine-tuning for domain-specific data, on-device AI applications, and rapid prototyping, offering strong performance in math reasoning and Python code generation. Its non-MoE architecture simplifies understanding and adaptation for various specialized tasks and languages.
Loading preview...
Jibay 2: A Customizable and Efficient Language Model
Jibay 2, developed by JibayAi, is a lightweight, open-source language model based on the Qwen 3 architecture. Released in Khordad 1405 (May/June 2026), it features 2 billion active parameters and a substantial 32,768 token context length.
Key Capabilities & Features
- Simple and Trainable: Designed for ease of understanding, debugging, and fine-tuning due to its non-MoE (Mixture of Experts) architecture, where all parameters are active.
- Efficient Performance: Offers a practical alternative for developers seeking a controllable model without high computational demands.
- Strong Benchmarks: Achieves notable scores including ~70.0% on GSM8K for math reasoning and ~68.0% on HumanEval for Python code generation.
- Customization Focus: Can be entirely retrained, extended, or adapted to various fields and languages, including medical/legal text processing, niche code generation, and low-resource language conversational AI.
- Optimized Usage: Recommends GGUF quantized versions for best performance, especially on consumer hardware.
Good For
- Fine-tuning on domain-specific datasets.
- On-device or edge AI applications.
- Research and educational purposes.
- Lightweight agentic workflows with function calling capabilities.
- Rapid prototyping and experimentation across diverse languages and tasks.