Perplexity Benchmarks
#7
by
thad0ctor
- opened
Would it be possible to post perplexity mesaurements please?
Would it be possible to post perplexity mesaurements please?
Perplexity is a poor measurement of quality for quants. KL Divergence is usually the answer but doing those benchmarks take a lot of time. In general, Q2_K_XL is always the best in terms of size + accuracy
There's aider polygot benchmark for Q4_K_XL: https://huggingface.co/unsloth/Qwen3-Coder-480B-A35B-Instruct-GGUF/discussions/8