LoTA-QAF: Lossless Ternary Adaptation for Quantization-Aware Fine-Tuning | Read Paper on Bytez