标签：#hack

"I wanted to see if I could optimize the dequantization bottleneck during 4-bit LLM inference. By writing a custom kernel in Triton to optimize memory access patterns, I managed to get up to a 1.41x speedup over the standard bitsandbytes implementation. Check out the source code and benchmarks, feedback is highly appreciated!"

I feel like I’m going crazy, I’ve been hitting the strangest bugs across all my devices in the last 6 months. Is anyone else experiencing this too?