[benchmark] Qwen3.5-35B-A3B-4bit on Apple M5 Max

[magicnight]

Benchmark data (auto-generated)

{
  "completionTokens" : 2023,
  "engineID" : "mlx-swift-lm",
  "engineVersion" : "mlx-swift-lm 3.31.3",
  "generationTPS" : 133.47814333534475,
  "id" : "800C438E-53B7-4983-90A6-4B1F9AAD74B2",
  "macMLXVersion" : "0.3.5",
  "memoryUsedGB" : 52.18634033203125,
  "modelID" : "Qwen3.5-35B-A3B-4bit",
  "modelLoadTimeS" : 0,
  "notes" : "",
  "promptTPS" : 2756.5669981190003,
  "promptTokens" : 1479,
  "runs" : 3,
  "system" : {
    "chip" : "Apple M5 Max",
    "macOSVersion" : "26.4.1",
    "ramGB" : 128
  },
  "timestamp" : "2026-04-17T18:36:15Z",
  "ttftMs" : 536.5369319915771
}

Optional context

No response