23:01
2026-09-26
pub.towardsai.net
large-language-models
Fine-Tuning to Quantization: What Free-Tier Hardware Can Prove
QLoRA fine-tuning on 150 synthetic examples raised a base Qwen3.5-4B model's adversarial-prompt refusal rate from 79.5% (31 of 39) to 89.7% (35 of 39), while GPTQ and AWQ 4-bit quantization added roug…