RYS-XLargeAfter testing several smaller models (Llama’s and smaller Qwen2’s), I set up the config for Qwen2-72B and let it sweep. Each $(i, j)$ configuration took a few minutes: load the re-layered model, run the math probe, run the EQ probe, record the scores, move on. Days of continuous GPU time on the 4090s. But far less compute than a fine tune! In fact, I didn’t even have the hardware needed for a LORA fine-tune on just 48GB of VRAM.
Consider two key arguments, both of which are true:
，推荐阅读91吃瓜获取更多信息
involved; instead of (struct.new $t (i32.const 42)), you use throw:
https://feedx.net