Add result: IQ3_M on Qwen2.5-1.5B-Instruct@e2cc88932f748bcb
#4
by vanishingradient - opened
Result for IQ3_M (hf://bartowski/Qwen2.5-1.5B-Instruct-GGUF/Qwen2.5-1.5B-Instruct-IQ3_M.gguf)
- pack:
Qwen2.5-1.5B-Instruct@e2cc88932f748bcb(ide2cc88932f748bcb) - quant file SHA-256:
aebd579aa34bde75426b7e3b786b089bc366f16da19e8aa60945d27f77e780f0 - backend: llamacpp (llama-cpp-python 0.3.36)
- hardware: Tesla T4 / CUDA (Linux)
- KLD mean: 0.14942 (95% CI 0.13453 to 0.16615)
- top-1 agreement: 82.62%
- approximate (server top-N): no
- loops: 21 (reference 16), wrong script: 0, valid JSON: 8/8 (reference 8)
To re-check: quantgap score <same file> --pack <same pack> should give a KLD mean within 2%.
Submitted with quantgap 0.1.1.
Merged by maintainer: Kaggle T4 full-mode run, sha256 verified against plan.
vanishingradient changed pull request status to merged