Downloads · 30 days
0
kld-baselines-gguf
krampenschiesser/kld-baselines-gguf
kld-baselines-gguf is a machine learning model from krampenschiesser. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
These are KLD files generated by llama.cpp's llama-perplexity -s 42 --kl-divergence-base myfile.kld -fa 1 -c 2048 -f wiki.test.raw -m mymodel.gguf
Access
Public
Updated Oct 22, 2026
Repo size
677 GB
Likes
1
Public
Repo makeup
Click a slice to open those files.
.kld635 GB · 94%
At a glance
- License
- mit
- Access
- Public
- Created
- Feb 26, 2026
- Updated
- Oct 22, 2026
- SHA
- c124cee3
Model card
From the Hugging Face model README
Llama.cpp KLD baselines based on wiki.test.raw
These are KLD files generated by llama.cpp's
llama-perplexity -s 42 --kl-divergence-base myfile.kld -fa 1 -c 2048 -f wiki.test.raw -m mymodel.gguf
When possible I used bf16 as baseline model. However due to limits on my VRAM I used the q8_0 quant for large models as baseline.
Use them for testing the KL divergence of your quants:
llama-perplexity --kl-divergence --kl-divergence-base myfile.kld -fa 1 -c 2048 -f wiki.test.raw -m mymodel.gguf
- License
- mit
- Created
- Feb 26, 2026
- Updated
- Oct 22, 2026