Add result: Q3_K_M on Qwen2.5-1.5B-Instruct@e2cc88932f748bcb

#7

Result for Q3_K_M (hf://Qwen/Qwen2.5-1.5B-Instruct-GGUF/qwen2.5-1.5b-instruct-q3_k_m.gguf)

  • pack: Qwen2.5-1.5B-Instruct@e2cc88932f748bcb (id e2cc88932f748bcb)
  • quant file SHA-256: 58cb5c05ecef48e82961f1a2be6544145ea26136f69dddda4bbbd092f0e4b993
  • backend: llamacpp (llama-cpp-python 0.3.36)
  • hardware: Tesla T4 / CUDA (Linux)
  • KLD mean: 0.15920 (95% CI 0.14251 to 0.17776)
  • top-1 agreement: 82.45%
  • approximate (server top-N): no
  • loops: 24 (reference 16), wrong script: 0, valid JSON: 6/8 (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

Sign up or log in to comment