{
  "benchmark_env": "openvino-2026.3.0-py312",
  "analysis_env": "openvino-3.12",
  "precision": "FP16",
  "models": [
    {"key": "qwen25_0_5b_fp16", "family": "Qwen2.5", "size_b": 0.5, "precision": "FP16", "directory": "qwen25_0_5b_ov_fp16"},
    {"key": "qwen25_1_5b_fp16", "family": "Qwen2.5", "size_b": 1.5, "precision": "FP16", "directory": "qwen25_1_5b_ov_fp16"},
    {"key": "qwen25_3b_fp16", "family": "Qwen2.5", "size_b": 3.0, "precision": "FP16", "directory": "qwen25_3b_ov_fp16"},
    {"key": "qwen25_7b_fp16", "family": "Qwen2.5", "size_b": 7.0, "precision": "FP16", "directory": "qwen25_7b_ov_fp16"},
    {"key": "llama32_1b_fp16", "family": "Llama 3.2", "size_b": 1.0, "precision": "FP16", "directory": "llama32_1b_ov_fp16"},
    {"key": "llama32_3b_fp16", "family": "Llama 3.2", "size_b": 3.0, "precision": "FP16", "directory": "llama32_3b_ov_fp16"}
  ],
  "contexts": [1024, 2048, 4096, 8192],
  "placements": ["openvino_auto", "gpu_gpu", "gpu_npu", "npu_gpu_combined"],
  "context_guard_tokens": 16,
  "output_tokens": 16,
  "repeats": 3,
  "teacher_token": 42,
  "warmup_prompt_tokens": 64,
  "cooldown_seconds": 2.0,
  "idle_baseline_seconds": 10.0
}