This qwen2 model was trained 2x faster with Unsloth
adapter_model.safetensors
ref/adapter_model.safetensors
unsloth/Qwen2.5-3B-Instruct-bnb-4bit