Q4_K_L metadata wrongly indicates Q4_K_M quantization
#1
by justinbowes - opened
Can't pull directly in llama.cpp. Inspecting the Q4_K_L GGUF, the quantization is shown as Q4_K_M. The filesizes are different.
- Q3_K_L: not affected
- Q4_K_L: as above
- Q5_K_L: file_type is Q5_K_M (same issue)
- Q6_K_L: file_type is Q6_K (similar issue)
I just spot-checked bartowski/Qwen_Qwen3.5-397B-A17B-GGUF and Q5 looks the same.
yeah that's just how it works, there is no "Q4_K_L" officially, it's Q4_K_M with input/output set to Q8_0