Weiver-Q1 is a compact and efficient variant of Qwen3-4B, designed for general coding and reasoning tasks under limited computational resources.
The model is trained using the OpenCodeReasoning dataset, which contains high-quality synthetic data distilled from DeepSeek-R1, covering a broad range of programming and logical reasoning scenarios.
Fine-tuning was performed using QLoRA, enabling a significant reduction in memory footprint while maintaining strong performance.
This release is provided as a GGUF INT4 (Q4_K_M) quantized model, making it well-suited for local inference, edge devices, and low-latency applications.
The model is released under the Apache 2.0 license, allowing commercial and non-commercial use.
