Benchmarks/Qwen3-30B-A3B-Instruct-2507-unsloth-IQ4_NL/bench_metrics.json

{
  "llamabench": {
    "backend": "CUDA",
    "log_path": "Qwen3-30B-A3B-Instruct-2507-unsloth-IQ4_NL/llamabench.md",
    "ngl": "35",
    "raw_row": {
      "backend": "CUDA",
      "model": "qwen3moe 30B.A3B IQ4_NL - 4.5 bpw",
      "ngl": "35",
      "params": "30.53 B",
      "size": "16.26 GiB",
      "t/s": "138.69 \u00b1 8.16",
      "test": "pp8",
      "tps_value": 138.69
    },
    "test": "pp8",
    "tps": 138.69
  },
  "perplexity": {
    "code": {
      "log_path": "Qwen3-30B-A3B-Instruct-2507-unsloth-IQ4_NL/perplexity_code.log",
      "ppl": 1.3111,
      "ppl_error": 0.00733
    },
    "general": {
      "log_path": "Qwen3-30B-A3B-Instruct-2507-unsloth-IQ4_NL/perplexity_general.log",
      "ppl": 6.2669,
      "ppl_error": 0.12736
    },
    "math": {
      "log_path": "Qwen3-30B-A3B-Instruct-2507-unsloth-IQ4_NL/perplexity_math.log",
      "ppl": 5.7159,
      "ppl_error": 0.10611
    }
  }
}