Add result: IQ3_M on SmolLM2-135M-Instruct@58c854541f94686f

#12

Result for IQ3_M (hf://bartowski/SmolLM2-135M-Instruct-GGUF/SmolLM2-135M-Instruct-IQ3_M.gguf)

  • pack: SmolLM2-135M-Instruct@58c854541f94686f (id 58c854541f94686f)
  • quant file SHA-256: e8a0e942fe93529b7601a7620c8aa1bdc1bede40ff668ed9d6339686384392d7
  • backend: llamacpp (llama-cpp-python 0.3.37)
  • hardware: Tesla T4 / CUDA (Linux)
  • KLD mean: 0.12630 (95% CI 0.11847 to 0.13510)
  • top-1 agreement: 78.78%
  • approximate (server top-N): no
  • loops: 42 (reference 26), wrong script: 0, valid JSON: 3/8 (reference 2)

To re-check: quantgap score <same file> --pack <same pack> should give a KLD mean within 2%.

Submitted with quantgap 0.1.1.

Verified full-mode result (dry-run passed); merging.

vanishingradient changed pull request status to merged

Sign up or log in to comment