Files
ik_llama.cpp/github-data/pull_requests/329 - Add ability to hide imatrix details in llama-quantize.md
2025-07-23 13:31:53 +02:00

1.0 KiB

🔀 #329 - Add ability to hide imatrix details in llama-quantize

Author ikawrakow
State Closed
Created 2025-04-14
Updated 2025-04-14

Description

Simply add --hide-imatrix to the command line when quantizing. This will store "top_secret" in the imatrix data file name and calibration dataset fields, and zeros in the batch size and number of chunks used to compute the imatrix. Example:

llama_model_loader: - kv  29:                      quantize.imatrix.file str              = top_secret
llama_model_loader: - kv  30:                   quantize.imatrix.dataset str              = top_secret
llama_model_loader: - kv  31:             quantize.imatrix.entries_count i32              = 0
llama_model_loader: - kv  32:              quantize.imatrix.chunks_count i32              = 0

Why? Someone publishing quantized models may not want to reveal the details of the imatrix they have used.