{
  "artifacts": [
    {
      "file": "minicpm_wi4b32_wi8_afp32.litertlm",
      "sha256": "43f41a837c58408192d703a7cf7bb27a3f8b4f8cc1d34c7f79e54f058f966ccc",
      "size_mb": 755.906
    }
  ],
  "benchmarks": [],
  "conversion": {
    "command": "n/a (official artifact)",
    "quantization": "mixed INT4-block32 (linear) / INT8 (embed and lm_head) quantization (wi4b32_wi8) with FP32 activations (afp32) (README Available Models)",
    "tool": "official LiteRT-LM release of MiniCPM5-1B (README 'Available Models'); not converted in this lane",
    "tool_version": "n/a (official artifact)"
  },
  "cross_runtime": [],
  "delegation": null,
  "device": {
    "records": [
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "cpu",
          "context_length": null,
          "date": "2026-09-06",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max)",
            "machine_label": "mac-studio-m4-max",
            "os_build": null,
            "runtime": "litert-lm",
            "runtime_version": "0.17.0",
            "soc": null,
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "compat_check status ok (runtime /Users/USER/code/litertlm-convert/.qa-venvs/litert-lm-0.17.0/bin/litert-lm 0.17.0); fixed-question answer: '[thought] \\nWe are asked: \"What is 17 + 25? Answer briefly.\" So we need to compute the sum of 17 and 25. The sum is 42. We should answer briefly, so just \"42\" or maybe \"17+25=42\". But the instruction says \"Answer briefly.\" So we can simply say \"42\". I\\'ll answer: 42.\\n [/thought]\\n\\n\\n42'"
          ],
          "failure_class": null,
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/0.17.0/2026-09-06/minicpm5-1b__mac-studio-m4-max.json"
      }
    ]
  },
  "model": {
    "family": "minicpm5",
    "id": "minicpm5-1b",
    "license": "apache-2.0",
    "source_url": "https://huggingface.co/litert-community/MiniCPM5-1B",
    "task": "text-generation"
  },
  "pitfalls": [
    "The repo also ships minicpm_wi4b32_wi8_afp32_gpu_opt.litertlm (same recipe, optimized for GPU execution) and MiniCPM5-1B_dynamic_wi8_afp32.litertlm (dynamic weight-only INT8, static prefill memory allocation) — separate artifacts, not this card (README Available Models).",
    "Quantization benchmark in the README compares the FP (bf16) baseline and the W4 model with thinking mode turned off to reduce context usage (README Quantization Benchmark).",
    "The owner's MiniCPM5-2B conversion (cards minicpm5-2b-int4/-int8) follows this artifact's packaging (verbatim chat_template.jinja on the jinja path) — see minicpm52b_work/FINDINGS.md premises.",
    "License Apache-2.0, consistent with upstream openbmb/MiniCPM5-1B (README License).",
    "LiteRT-LM .litertlm bundle: LiteRT.js cannot run it, so delegation stays null and there is no browser block; device rows come from data/device_runs/."
  ],
  "schema_version": "1.2"
}