JustScriptzz/nexus-plus-v2
JustScriptzz/nexus-plus-v2 is a 4.05 billion parameter causal language model fine-tuned from Qwen3-4B-Base. Developed by JustScriptzz, it was instruction-tuned using QLoRA on approximately 50,000 instruction examples. This model is optimized for general instruction-following tasks, serving as a capable base for various conversational and text generation applications. Its compact size and fine-tuning approach make it suitable for experimentation and further domain-specific adaptation.
Loading preview...
Nexus Plus v2 Overview
JustScriptzz/nexus-plus-v2 is a 4.05 billion parameter causal language model built upon the Qwen3-4B-Base architecture. It has been instruction-tuned using QLoRA on a dataset comprising approximately 50,000 instruction examples, including Dolly-15k, synthetic QA, and general instruction data. This fine-tuning process involved 33 million trainable parameters with a LoRA rank of 16 and alpha of 32, targeting the q_proj and v_proj modules.
Key Capabilities & Features
- Instruction Following: Fine-tuned to understand and respond to a variety of instructions.
- Efficient Fine-tuning: Utilizes QLoRA for efficient adaptation from its base model.
- Merged Model Availability: Provided as a fully merged model, eliminating the need for PEFT adapters during inference.
- Accessible Hardware: Trained on consumer-grade hardware (RTX 5060 Ti 16GB) in about 7 hours.
Use Cases & Limitations
This model is well-suited for general text generation and conversational tasks where a smaller, efficient model is preferred. It can serve as an excellent starting point for developers looking to experiment with instruction-tuned models or for further fine-tuning on specific datasets. However, due to its relatively small training dataset, it may not generalize perfectly across all domains and is best considered for learning experiments or as a foundational model for more specialized applications.