mtplx_pair.json

{
  "format_version": 1,
  "name": "Gemma4 MTPLX Optimized Speed",
  "variant": "speed",
  "layout": {
    "target": "target",
    "assistant": "assistant"
  },
  "source": {
    "target_repo": "google/gemma-4-31B-it",
    "target_revision": "145dc2508c480a64b47242f160d286cff94a2343",
    "assistant_repo": "google/gemma-4-31B-it-assistant",
    "assistant_revision": "cffbbd2cea41ea56a0fa5b0487e0d445121fd204"
  },
  "target": {
    "role": "verifier",
    "model_type": "gemma4",
    "quantization": {
      "bits": 4,
      "group_size": 64,
      "mode": "affine"
    }
  },
  "assistant": {
    "role": "drafter",
    "model_type": "gemma4_assistant",
    "quantization": {
      "bits": 6,
      "group_size": 64,
      "mode": "affine"
    }
  },
  "benchmark": {
    "prompt_suite": "flappy",
    "max_tokens": 1000,
    "temperature": 1.0,
    "top_p": 0.95,
    "top_k": 64,
    "seed": 0,
    "best_block_size": 6,
    "acceptance": {
      "accepted": 830,
      "drafted": 846,
      "ratio": 0.9810874704491725
    },
    "observed_mtp_tok_s": [
      43.558797060869416,
      44.457398919489286,
      44.071735166669846
    ]
  }
}