{
  "schema_version": 1,
  "measured_at_utc": "2026-10-28T04:03:38.308408+00:00",
  "hardware": {
    "gpu": "NVIDIA GeForce RTX 4060 Laptop GPU",
    "memory_total_mib": 8188.0,
    "driver": "592.82"
  },
  "ollama_version": "0.40.11",
  "synthetic_task": "Integer parity with strict JSON labels; no personal inputs",
  "runs": [
    {
      "runtime": "Ollama",
      "model": "qwen2.5:0.5b-instruct",
      "model_digest": "a8b0c51577010a279d933d14c2a8ab4b268079d44c5c8830c0a93900f1827c67",
      "quantization": "Q4_K_M",
      "parameters": "494.03M",
      "samples": 24,
      "warmup_requests_excluded": 1,
      "context_limit": 2048,
      "output_limit": 32,
      "seed": 42,
      "temperature": 0,
      "concurrency": 1,
      "gpu_resident_bytes_after_warmup": 456382217,
      "json_valid": 24,
      "correct": 12,
      "p50_wall_s": 0.13749715000449214,
      "p95_wall_s": 0.16817859999719076,
      "output_tokens": 240,
      "generation_duration_s": 1.193776997,
      "output_tokens_per_s": 201.0425737831502
    },
    {
      "runtime": "Ollama",
      "model": "qwen2.5:1.5b",
      "model_digest": "65ec06548149b04c096a120e4a6da9d4017ea809c91734ea5631e89f96ddc57b",
      "quantization": "Q4_K_M",
      "parameters": "1.5B",
      "samples": 24,
      "warmup_requests_excluded": 1,
      "context_limit": 2048,
      "output_limit": 32,
      "seed": 42,
      "temperature": 0,
      "concurrency": 1,
      "gpu_resident_bytes_after_warmup": 1108565032,
      "json_valid": 24,
      "correct": 13,
      "p50_wall_s": 0.16591280000284314,
      "p95_wall_s": 0.18953280002460815,
      "output_tokens": 240,
      "generation_duration_s": 1.872521,
      "output_tokens_per_s": 128.16945711156242
    }
  ],
  "limitations": [
    "Single laptop and single trial per model",
    "Short synthetic parity task does not measure general intelligence",
    "Different model sizes are not a controlled quantization comparison",
    "Wall latency includes complete generation, not time to first token",
    "GPU residency is runtime-reported after warmup, not peak device memory"
  ],
  "training_status": "completed",
  "training": {
    "method": "LoRA FP16 frozen base; not QLoRA",
    "runtime": "Transformers/PyTorch CUDA",
    "model": "Qwen/Qwen2.5-0.5B-Instruct",
    "model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
    "model_license": "Apache-2.0",
    "seed": 42,
    "training_examples": 32,
    "training_steps": 32,
    "optimizer_steps": 16,
    "microbatch": 1,
    "gradient_accumulation": 2,
    "sequence_limit": 256,
    "adapter_rank": 4,
    "trainable_parameters": 270336,
    "total_parameters": 494303104,
    "training_elapsed_s": 11.307847000018228,
    "peak_torch_allocated_mib": 1070.95947265625,
    "initial_training_loss": 0.6269833445549011,
    "final_training_loss": 0.3555484712123871,
    "baseline": {
      "samples": 12,
      "correct": 6,
      "accuracy": 0.5,
      "mean_latency_s": 0.38881636667550384
    },
    "adapted": {
      "samples": 12,
      "correct": 6,
      "accuracy": 0.5,
      "mean_latency_s": 0.2688223000053161
    },
    "versions": {
      "torch": "2.9.1+cu128",
      "transformers": "4.57.1",
      "peft": "0.17.1",
      "accelerate": "1.11.0"
    },
    "task": "Synthetic integer parity; training values 10-41, held-out values 200-211; same instruction template",
    "limitations": [
      "Small held-out set and shared prompt template; not broad generalization evidence",
      "One run and no high-precision versus quantized comparison",
      "Latency includes each complete generation; not time to first token"
    ]
  }
}