Benchmarks/Qwen3-30B-A3B-Instruct-2507-unsloth-Q8_0/bench_metrics.json

{
  "llamabench": {
    "backend": "CUDA",
    "log_path": "Qwen3-30B-A3B-Instruct-2507-unsloth-Q8_0/llamabench.md",
    "ngl": "35",
    "raw_row": {
      "backend": "CUDA",
      "model": "qwen3moe 30B.A3B Q8_0",
      "ngl": "35",
      "params": "30.53 B",
      "size": "30.25 GiB",
      "t/s": "95.03 \u00b1 2.99",
      "test": "pp8",
      "tps_value": 95.03
    },
    "test": "pp8",
    "tps": 95.03
  },
  "perplexity": {
    "code": {
      "log_path": "Qwen3-30B-A3B-Instruct-2507-unsloth-Q8_0/perplexity_code.log",
      "ppl": 1.2991,
      "ppl_error": 0.00723
    },
    "general": {
      "log_path": "Qwen3-30B-A3B-Instruct-2507-unsloth-Q8_0/perplexity_general.log",
      "ppl": 6.2536,
      "ppl_error": 0.12771
    },
    "math": {
      "log_path": "Qwen3-30B-A3B-Instruct-2507-unsloth-Q8_0/perplexity_math.log",
      "ppl": 5.7045,
      "ppl_error": 0.10628
    }
  }
}