{
  "schema_version": 1,
  "date_utc": "2026-09-27",
  "model_id": "matteiuspi/Qwen3.8-Flash-Next-W4A16-G128-300i",
  "hardware": "2 x Atlas 300I Duo / 4 x Ascend 310P3",
  "checkpoint_format": "W4A16 G128, packed INT4 routed experts, FP16 activations",
  "runtime": "experimental W4 tile-reuse candidate, MTP draft depth 2, FULL_DECODE_ONLY graphs, TP4/EP",
  "sampling": {
    "temperature": 0,
    "seed": 42,
    "thinking": false,
    "ignore_eos": true,
    "completion_tokens": 512
  },
  "metric": "(completion_tokens - 1) / (elapsed_seconds - time_to_first_token_seconds), based on client streaming; excludes prefill",
  "workloads": [
    {
      "label": "short_prompt",
      "source_run": "tile-http-candidate-short.jsonl",
      "measured_requests": [
        {
          "prompt_tokens": 32,
          "cached_prompt_tokens": 0,
          "completion_tokens": 512,
          "time_to_first_token_seconds": 0.5066101450065617,
          "decode_seconds": 31.007059219999064,
          "client_decode_tokens_per_second": 16.480118168394796,
          "mtp_acceptance": 0.8643617021276596,
          "output_sha256": "05ab57b2bc71b7bd6efb1f43d5ed54885de933122a4120e11373ed6258f95ef8"
        },
        {
          "prompt_tokens": 43,
          "cached_prompt_tokens": 0,
          "completion_tokens": 512,
          "time_to_first_token_seconds": 0.6068624640029157,
          "decode_seconds": 36.076975575997494,
          "client_decode_tokens_per_second": 14.164158492819318,
          "mtp_acceptance": 0.6820276497695853,
          "output_sha256": "d6f9cf24f10f2b6fcf8c9664648ce6d499ff2e226bbb78a9f6e4791d64d26d17"
        },
        {
          "prompt_tokens": 49,
          "cached_prompt_tokens": 0,
          "completion_tokens": 512,
          "time_to_first_token_seconds": 0.6193453000014415,
          "decode_seconds": 34.727616406999005,
          "client_decode_tokens_per_second": 14.714514063136594,
          "mtp_acceptance": 0.7166666666666667,
          "output_sha256": "6f72caf6e5abf1b1dc3e370968ad7cb683667a6597b86d5bf55989ffce2d6af0"
        }
      ],
      "median_client_decode_tokens_per_second": 14.714514063136594
    },
    {
      "label": "approximately_23k_prompt",
      "source_run": "tile-http-candidate-long.jsonl",
      "measured_requests": [
        {
          "prompt_tokens": 23407,
          "cached_prompt_tokens": 23168,
          "completion_tokens": 512,
          "time_to_first_token_seconds": 1.6225465909956256,
          "decode_seconds": 31.515383977006422,
          "client_decode_tokens_per_second": 16.2143034770836,
          "mtp_acceptance": 0.8864864864864865,
          "output_sha256": "cef47d1a306d8ad208bcf04501012707e5c2389d130dbc4a8e9d404662b23ef9"
        },
        {
          "prompt_tokens": 23418,
          "cached_prompt_tokens": 23168,
          "completion_tokens": 512,
          "time_to_first_token_seconds": 1.6728898490036954,
          "decode_seconds": 34.99341846400057,
          "client_decode_tokens_per_second": 14.602745957091633,
          "mtp_acceptance": 0.7524509803921569,
          "output_sha256": "9b1f608c46be1e37542ec02ded7c803cc061326094ad6ad3c59a5c84cbcb53ad"
        },
        {
          "prompt_tokens": 23424,
          "cached_prompt_tokens": 23168,
          "completion_tokens": 512,
          "time_to_first_token_seconds": 2.3089601309984573,
          "decode_seconds": 36.66200596400449,
          "client_decode_tokens_per_second": 13.938135313755343,
          "mtp_acceptance": 0.6985981308411215,
          "output_sha256": "6866fdd0b999ca2fc1a513eb69363ec95bd9beb8c20da37d8f232c769a70976c"
        }
      ],
      "median_client_decode_tokens_per_second": 14.602745957091633
    }
  ],
  "scope": "Three requests per workload in one W4 runtime study. W8 was not rerun on this configuration. Fixed-length outputs and prompt checks do not establish full-model quality or 262K context correctness."
}