{
  "date": "2026-10-25",
  "hardware": {
    "gpu": "AMD Radeon AI PRO R9700",
    "gpu_vram_gib": 31.86,
    "vm_ram_gib": 31.3
  },
  "configuration": {
    "context_tokens_per_slot": 250000,
    "parallel": 2,
    "kv": "int8",
    "kv_resident_tokens": 32768,
    "mtp_draft": false,
    "vision": false,
    "engine_version": "0.1.48"
  },
  "model": {
    "repository": "agentionai/Qwen3.8-Flash-Next-Gyro-abliterated-Huihui-GGUF",
    "filename": "Qwen3.8-Flash-Next-Gyro-S-abliterated-TQ1_0.gguf",
    "sha256": "783e1af72859c48146162840f1a2173922c03e4befaeea4add7578ad25c303a3",
    "verified": true
  },
  "methodology": {
    "measured_runs_per_concurrency": 1,
    "tested_concurrency": [
      1,
      2,
      4
    ],
    "not_tested": [
      3
    ],
    "long_prompt_tokens_per_request": 248894,
    "long_max_output_tokens": 768,
    "short_prompt_tokens_per_request": 92,
    "short_max_output_tokens": 256,
    "temperature": 0,
    "time_to_first_output": "First streamed content or reasoning token; excludes model startup.",
    "generation_rate": "Completion tokens include reasoning. Generation interval can include other requests prefill pauses.",
    "retrieval_test": "Synthetic numbered archive; return three known target values near beginning, middle, and end.",
    "limitations": "One synthetic retrieval prompt, about 249K input tokens per user. One measured run per concurrency. Three users were not tested. Generation rates count reasoning tokens and include pauses caused by other requests. More RAM has not been tested."
  },
  "successful_trials": [
    {
      "requested_slots": 1,
      "short": {
        "wall_seconds": 4.2687060330063105,
        "aggregate_tokens_per_second": 59.971335111991195,
        "requests": [
          {
            "request": 0,
            "seconds": 4.266429285984486,
            "ttft_seconds": 0.5217867596074939,
            "usage": {
              "prompt_tokens": 92,
              "completion_tokens": 256,
              "total_tokens": 348,
              "prompt_tokens_details": {
                "cached_tokens": 0
              }
            },
            "end_to_end_tokens_per_second": 60.00333835157601,
            "decode_tokens_per_second": 68.09724968022697,
            "finish_reason": "length"
          }
        ]
      },
      "249k": {
        "wall_seconds": 109.5865449118428,
        "aggregate_tokens_per_second": 5.009739110231504,
        "requests": [
          {
            "request": 0,
            "seconds": 109.58562790881842,
            "ttft_seconds": 102.185088314116,
            "usage": {
              "prompt_tokens": 248894,
              "completion_tokens": 549,
              "total_tokens": 249443,
              "prompt_tokens_details": {
                "cached_tokens": 0
              }
            },
            "end_to_end_tokens_per_second": 5.009781031293627,
            "decode_tokens_per_second": 74.04863433305988,
            "finish_reason": "stop",
            "retrieval_pass": true
          }
        ]
      }
    },
    {
      "requested_slots": 2,
      "short": {
        "wall_seconds": 7.0010602958500385,
        "aggregate_tokens_per_second": 73.13177981105149,
        "requests": [
          {
            "request": 0,
            "seconds": 6.957969688810408,
            "ttft_seconds": 0.510802845004946,
            "usage": {
              "prompt_tokens": 92,
              "completion_tokens": 256,
              "total_tokens": 348,
              "prompt_tokens_details": {
                "cached_tokens": 0
              }
            },
            "end_to_end_tokens_per_second": 36.7923419401627,
            "decode_tokens_per_second": 39.55223879080243,
            "finish_reason": "length"
          },
          {
            "request": 1,
            "seconds": 6.998851340729743,
            "ttft_seconds": 1.0828254716470838,
            "usage": {
              "prompt_tokens": 92,
              "completion_tokens": 256,
              "total_tokens": 348,
              "prompt_tokens_details": {
                "cached_tokens": 0
              }
            },
            "end_to_end_tokens_per_second": 36.57743071498185,
            "decode_tokens_per_second": 43.103245370944975,
            "finish_reason": "length"
          }
        ]
      },
      "249k": {
        "wall_seconds": 241.52554543921724,
        "aggregate_tokens_per_second": 4.359787276683741,
        "requests": [
          {
            "request": 0,
            "seconds": 241.5243949200958,
            "ttft_seconds": 230.174455948174,
            "usage": {
              "prompt_tokens": 248894,
              "completion_tokens": 549,
              "total_tokens": 249443,
              "prompt_tokens_details": {
                "cached_tokens": 0
              }
            },
            "end_to_end_tokens_per_second": 2.273062313981274,
            "decode_tokens_per_second": 48.282188445147085,
            "finish_reason": "stop",
            "retrieval_pass": true
          },
          {
            "request": 1,
            "seconds": 179.4108332511969,
            "ttft_seconds": 103.14177091186866,
            "usage": {
              "prompt_tokens": 248894,
              "completion_tokens": 504,
              "total_tokens": 249398,
              "prompt_tokens_details": {
                "cached_tokens": 0
              }
            },
            "end_to_end_tokens_per_second": 2.8091949124072064,
            "decode_tokens_per_second": 6.595072293529312,
            "finish_reason": "stop",
            "retrieval_pass": true
          }
        ]
      }
    }
  ],
  "four_user_trial": {
    "concurrency": 4,
    "short_requests": {
      "concurrency": 4,
      "wall_seconds": 13.451596293132752,
      "aggregate_tokens_per_second": 76.1247942389386,
      "results": [
        {
          "request": 0,
          "seconds": 13.448746969923377,
          "ttft_seconds": 1.6194804888218641,
          "usage": {
            "prompt_tokens": 92,
            "completion_tokens": 256,
            "total_tokens": 348,
            "prompt_tokens_details": {
              "cached_tokens": 0
            }
          },
          "end_to_end_tokens_per_second": 19.03523061089003,
          "decode_tokens_per_second": 21.556701325676052,
          "finish_reason": "length"
        },
        {
          "request": 1,
          "seconds": 13.448724957648665,
          "ttft_seconds": 2.1004354269243777,
          "usage": {
            "prompt_tokens": 92,
            "completion_tokens": 256,
            "total_tokens": 348,
            "prompt_tokens_details": {
              "cached_tokens": 0
            }
          },
          "end_to_end_tokens_per_second": 19.035261766908665,
          "decode_tokens_per_second": 22.47034473899114,
          "finish_reason": "length"
        },
        {
          "request": 2,
          "seconds": 13.35520125599578,
          "ttft_seconds": 0.5557707571424544,
          "usage": {
            "prompt_tokens": 92,
            "completion_tokens": 256,
            "total_tokens": 348,
            "prompt_tokens_details": {
              "cached_tokens": 0
            }
          },
          "end_to_end_tokens_per_second": 19.168561753052543,
          "decode_tokens_per_second": 19.92275821837218,
          "finish_reason": "length"
        },
        {
          "request": 3,
          "seconds": 13.448402178939432,
          "ttft_seconds": 1.1465336089022458,
          "usage": {
            "prompt_tokens": 92,
            "completion_tokens": 256,
            "total_tokens": 348,
            "prompt_tokens_details": {
              "cached_tokens": 0
            }
          },
          "end_to_end_tokens_per_second": 19.03571863733396,
          "decode_tokens_per_second": 20.728557215032325,
          "finish_reason": "length"
        }
      ]
    },
    "long_requests": {
      "concurrency": 4,
      "wall_seconds": 484.60054663335904,
      "aggregate_tokens_per_second": 0.0,
      "results": [
        {
          "request": 0,
          "seconds": 483.3303927988745,
          "ttft_seconds": null,
          "usage": {},
          "end_to_end_tokens_per_second": null,
          "decode_tokens_per_second": null,
          "finish_reason": null
        },
        {
          "request": 1,
          "seconds": 484.59738148003817,
          "ttft_seconds": null,
          "usage": {},
          "end_to_end_tokens_per_second": null,
          "decode_tokens_per_second": null,
          "finish_reason": null
        },
        {
          "request": 2,
          "seconds": 483.3270048862323,
          "ttft_seconds": null,
          "usage": {},
          "end_to_end_tokens_per_second": null,
          "decode_tokens_per_second": null,
          "finish_reason": null
        },
        {
          "request": 3,
          "seconds": 483.32541922386736,
          "ttft_seconds": null,
          "usage": {},
          "end_to_end_tokens_per_second": null,
          "decode_tokens_per_second": null,
          "finish_reason": null
        }
      ]
    },
    "cause": "Engine stopped itself after 60 seconds without progress during batched prompt processing near token 248889. This is an observed runtime hang, not proof of a hard RAM or VRAM capacity limit."
  },
  "previous_35b_baseline": {
    "concurrency": 1,
    "wall_seconds": 566.950022675097,
    "aggregate_tokens_per_second": 0.4515389183549012,
    "results": [
      {
        "request": 0,
        "seconds": 566.9489130601287,
        "ttft_seconds": 562.9925382710062,
        "usage": {
          "completion_tokens": 256,
          "prompt_tokens": 248852,
          "total_tokens": 249108,
          "prompt_tokens_details": {
            "cached_tokens": 0
          }
        },
        "end_to_end_tokens_per_second": 0.4515398020929789,
        "decode_tokens_per_second": 64.45289444163325,
        "finish_reason": "length"
      }
    ]
  },
  "baseline_limitations": "Same archive text, different model templates and runtimes; old model used max_tokens=256 rather than 768 and its final answer was truncated. First-output comparison only.",
  "serving_concurrency_after_test": 2
}