From 100toolkit · · 50s
Qwen3-4B with writing tics trained out
AI model Qwen3-4B writes clean prose without AI tics after training.
In brief
AI model Qwen3-4B writes clean prose without AI tics after training. Qwen3-4B model trained to write human-like prose without AI writing patterns. Originally reported by 100toolkit.
Model setup
The quantised build runs on a laptop with Ollama or llama.cpp.
Full-precision weights
Full-precision weights load with transformers like any Qwen3-4B checkpoint.
LoRA adapter
The model includes a 126 MB LoRA adapter trained with the edit active.
Writing style
The model writes short declarative sentences and diary-style personal pieces.
Limitations
The model invents names and places but runs long and scores inconsistently.
Next steps
The creator plans to train a base model on public essays and letters for better writing.
In short
Qwen3-4B model trained to write human-like prose without AI writing patterns.
