User Seeks Hardware Advice for Full Fine-Tuning Qwen3.5-14B

A user plans to fully fine-tune Qwen3.5-14B-Base on about 18,000 long examples that conflict with the model's knowledge, ruling out LoRA. After repeatedly hitting out-of-memory errors on a vast.ai machine with four 12GB GPUs and 48GB RAM, wasting about $3, the user is seeking hardware recommendations.