{
  "artifacts": [
    {
      "file": "mnn-pocket-0.3b-dec_L512_fp16w_npu_SM8850.tflite",
      "sha256": "31d729102125cd2214e9e647c15efae627f1bdb0262c74a03c7a953b201a0583",
      "size_mb": 485.863
    }
  ],
  "benchmarks": [],
  "conversion": {
    "command": "the repository's conversion/ (conversion/README.md) holds the scripts and REPRODUCE.md every check (README L150); the card names the tools, not a command line",
    "quantization": "float16 linear-layer weights, the LoRA adapter folded into the weights first; SiLU written with EXP in place of LOGISTIC for the NPU; compiled for the SM8850 ahead of time (README L20, L80, L129)",
    "tool": "litert-torch",
    "tool_version": "0.10.0 (for LiteRT 2.3.0); compiled for the SM8850 with QAIRT 2.50"
  },
  "cross_runtime": [],
  "delegation": null,
  "device": {
    "records": [
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "npu_qnn",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": 2600,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-mnnpocketdec-capi-runner",
            "os_build": "Android 16 (BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": "QAIRT 2.50.0.260828221209"
          },
          "error": null,
          "evidence": [
            "(tag s26_a_L512, run r4): run DONE rc 0, start 2026-11-07 21:22:18, 142 questions",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final-xsilu_L512_w16_SM8850.tflite rc=0 bytes 509464600/509464600 sha256 phone=31d729102125cd2214e9e647c15efae627f1bdb0262c74a03c7a953b201a0583 = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 12/12 OK",
            "Device: SM-S942Q 16 BP4A.251205.006.S942QOPS1AZH9",
            "[\"512\"]: the SM8850 compile on the Mac: plugin rc 0, partition selected 2600 of 2600 ops in 1 partition, SDK v2.50.0.260828221209, output sha256 = the Hub LFS sha256; on the phone runner.replacing: VERBOSE: Replacing 1 out of 1 node(s) with delegate (DispatchDelegate) node, yielding 1 partitions for subgraph 0 (main).",
            "Context binary SDK version matches current SDK: 2.50.0",
            "line PASS: 142 questions, same winner 137/137 (not near ties), near ties kept 3/5, max |Δp| 0.00599424, mean |Δp| 0.00111224, control n/a (the file holds no request of the control pair), non-finite read-outs 0, winner = the Mac CPU's on the same file 137/137",
            "Run memory (2 samples): VmHWM_max_kB 557960 VmSwap_max_kB 0 MemAvailable_min_kB 6233132 kgsl_page_alloc_max_B 214831104"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 15.0,
            "arm_files_run": 15.0,
            "compile_ms": 319.9,
            "line_bar_control_abs_dp": 0.02,
            "line_bar_max_abs_dp": 0.02,
            "line_bar_mean_abs_dp": 0.002,
            "line_bar_near_tie_gap": 0.02,
            "line_mac_same_file_non_near_tie": 137.0,
            "line_max_abs_dp": 0.005994238213396952,
            "line_mean_abs_dp": 0.0011122389769630137,
            "line_near_tie": 5.0,
            "line_non_near_tie": 137.0,
            "line_nonfinite_questions": 0.0,
            "line_pass": 1.0,
            "line_questions": 142.0,
            "line_winner_equal_mac_non_near_tie": 137.0,
            "line_winner_equal_near_tie": 3.0,
            "line_winner_equal_non_near_tie": 137.0,
            "vmhwm_max_kb": 557960.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the line: 142 questions of the L512 set against the source model's reference; the C-API runner build 7e7539e5 (LiteRT 2.3.0), Qualcomm HTP, the SM8850 context compiled ahead of time on the Mac, HTP performance mode Burst",
          "total_ops": 2600,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l512_fp16w_npu_sm8850__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "npu_qnn",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": 2600,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-mnnpocketdec-capi-runner",
            "os_build": "Android 16 (BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": "QAIRT 2.50.0.260828221209"
          },
          "error": null,
          "evidence": [
            "98.28 ms (min 96.92, max 102.05; median of 20 after 5 warm-ups, write+run+readback; run only 96.98 ms; load 0.33 s; VmHW… (the cell's value; parity cell: PASS; gate PASS, timing run = check run (4/4 q, dumps equal))",
            "(tag s26_at_b_L512, run r4): start 2026-11-07 21:34:12, request S/b_default (85 tokens), write_run_read_ms median 98.275 min 96.922 max 102.048 n 20, run_only_ms median 96.9825",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final-xsilu_L512_w16_SM8850.tflite rc=0 bytes 509464600/509464600 sha256 phone=31d729102125cd2214e9e647c15efae627f1bdb0262c74a03c7a953b201a0583 = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 12/12 OK; extras: rest=60 needs=s26_a_L512",
            "Device: SM-S942Q 16 BP4A.251205.006.S942QOPS1AZH9",
            "clock samples: 2 samples, 0 capped, GPU level max 0",
            "parity: [\"s26_at_b_L512\"]: gate PASS, timing run = check run (4/4 questions equal in every digit, max |Δp| between 0.0), against the check run",
            "VERBOSE: Replacing 1 out of 1 node(s) with delegate (DispatchDelegate) node, yielding 1 partitions for subgraph 0 (main)."
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 98.275,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 15.0,
            "arm_files_run": 15.0,
            "capped": 0.0,
            "capped_samples": 0.0,
            "clock_samples": 2.0,
            "compile_ms": 326.2,
            "latency_max_ms": 102.048,
            "latency_min_ms": 96.922,
            "request_questions": 3.0,
            "request_tokens": 85.0,
            "run_only_p50_ms": 96.9825,
            "timed_calls": 20.0,
            "vmhwm_max_kb": 558120.0,
            "warmup_calls": 5.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "timing: `default_request` (inference.py's DEFAULT_REQUEST) (85 tokens, 3 questions), the median of 20 calls after 5 warm-up calls in one process, a call = write the four inputs, run, read `hidden` back; the C-API runner build 7e7539e5 (LiteRT 2.3.0), Qualcomm HTP, the SM8850 context compiled ahead of time on the Mac, HTP performance mode Burst",
          "total_ops": 2600,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l512_fp16w_npu_sm8850__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "npu_qnn",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": 2600,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-mnnpocketdec-capi-runner",
            "os_build": "Android 16 (BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": "QAIRT 2.50.0.260828221209"
          },
          "error": null,
          "evidence": [
            "100.35 ms (min 99.02, max 102.18; median of 20 after 5 warm-ups, write+run+readback; run only 98.90 ms; load 0.33 s; VmH… (the cell's value; parity cell: PASS; gate PASS, timing run = check run (1/1 q, dumps equal))",
            "(tag s26_at_a_L512, run r4): start 2026-11-07 21:33:06, request S/a_readme (60 tokens), write_run_read_ms median 100.3455 min 99.019 max 102.178 n 20, run_only_ms median 98.9025",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final-xsilu_L512_w16_SM8850.tflite rc=0 bytes 509464600/509464600 sha256 phone=31d729102125cd2214e9e647c15efae627f1bdb0262c74a03c7a953b201a0583 = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 12/12 OK; extras: rest=60 needs=s26_a_L512",
            "Device: SM-S942Q 16 BP4A.251205.006.S942QOPS1AZH9",
            "clock samples: 2 samples, 0 capped, GPU level max 0",
            "parity: [\"s26_at_a_L512\"]: gate PASS, timing run = check run (1/1 questions equal in every digit, max |Δp| between 0.0), against the check run",
            "VERBOSE: Replacing 1 out of 1 node(s) with delegate (DispatchDelegate) node, yielding 1 partitions for subgraph 0 (main)."
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 100.3455,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 15.0,
            "arm_files_run": 15.0,
            "capped": 0.0,
            "capped_samples": 0.0,
            "clock_samples": 2.0,
            "compile_ms": 330.2,
            "latency_max_ms": 102.178,
            "latency_min_ms": 99.019,
            "request_questions": 1.0,
            "request_tokens": 60.0,
            "run_only_p50_ms": 98.9025,
            "timed_calls": 20.0,
            "vmhwm_max_kb": 558520.0,
            "warmup_calls": 5.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "timing: `readme_request` (the source model card's example) (60 tokens, 1 question), the median of 20 calls after 5 warm-up calls in one process, a call = write the four inputs, run, read `hidden` back; the C-API runner build 7e7539e5 (LiteRT 2.3.0), Qualcomm HTP, the SM8850 context compiled ahead of time on the Mac, HTP performance mode Burst",
          "total_ops": 2600,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l512_fp16w_npu_sm8850__galaxy-s26.json"
      }
    ]
  },
  "model": {
    "family": "mnn-pocket",
    "id": "mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l512_fp16w_npu_sm8850",
    "license": "apache-2.0",
    "source_url": "https://huggingface.co/litert-community/MNN-Pocket-0.3B-Dec-LiteRT",
    "task": "text-classification"
  },
  "pitfalls": [
    "A decision model, not a chat model: it takes a `state` and typed `questions` (`choice`, `noul`, `score`) and returns the probability of every candidate in one forward pass; it never generates text (README L18).",
    "The graph is the model body only: it takes the embeddings of the request's tokens, the rotary tables and the attention mask and returns the hidden states after the final norm; the Python host in `host/` builds the request, takes the embeddings from `embedding.safetensors` (the source model's token table) and reads the answers with the source model's head and temperature, as the source model's `inference.py` does (README L20, L147-L148).",
    "Each file is a static window: `<n>` in the name is the longest request it takes; run the smallest file that holds the request. A request longer than 2,048 tokens raises `ValueError` in the host, which never cuts it; the source model's own limit is 32,768 tokens (README L71, L84).",
    "The checks measure agreement with the source model's `inference.py`, unchanged, on its own files (float32, CPU, one thread): 2,409 questions on the Mac, a fixed subset of 16 to 322 questions per file on the phones; the bar is the same winner on every question whose two most likely options are more than 0.02 apart, a largest |dp| of at most 0.02 and a mean of at most 0.002 (README L88, L102).",
    "The repository has no Android host yet: an Android app has to port `host/pocket_dec_litert.py`; the phones ran LiteRT 2.3.0 through its C API with inputs the Mac computed with the host's code (README L137).",
    "Set the HTP performance mode to Burst: without it the calls on the Galaxy S26 took 5.5 to 6.7 times as long (76.3 ms against 11.3 ms for the 60-token example); in the C API it is `htp_performance_mode = 2` (`kLiteRtQualcommHtpPerformanceModeBurst`) in the Qualcomm options, in Kotlin `CompiledModel.QualcommOptions(htpPerformanceMode = CompiledModel.QualcommOptions.HtpPerformanceMode.BURST)` set as `qualcommOptions` of `CompiledModel.Options` (README L127).",
    "Compiled ahead of time for the Qualcomm SM8850 (Galaxy S26) with QAIRT 2.50: on the Galaxy S26 a file loads in 0.2 s (64 tokens) to 0.5 s (1,024 tokens) and 1.2 s (2,048 tokens), and the process peaks at 0.55 to 0.70 GB; for a Qualcomm NPU compiled on the phone the repository has the `_npu` files (README L80-L81, L129).",
    "SiLU is written with EXP in place of LOGISTIC, because the NPU's LOGISTIC returned 0 for inputs below about -5.7 (measured on the Galaxy S26 with QAIRT 2.47, on another model's graph) (README L129).",
    "Running it needs LiteRT's Qualcomm NPU runtime (Maven `litert-npu-runtime-qualcomm` 2.3.0) and Qualcomm's QAIRT 2.50 libraries; every file ran on the NPU in one partition, with all 2,600 ops (README L129).",
    "The Qualcomm NPU passes the bar (every winner that is not a near tie, 42 of 54 near ties, largest |dp| 0.0190) but can flip a near tie: in the source model's default example (`department`: `shipping` 0.41064 against `returns` 0.40709) it answers `returns`; from 64 to 1,024 tokens the files compiled ahead of time and the files compiled on the phone gave the same probabilities to every digit (README L97, L102, L104).",
    "The repository's scripts rebuild the 20 graphs other than the SM8850 files byte for byte on the Mac that made them; compiled again, the 64- and 768-token SM8850 files were the same computation, not the same bytes (README L150)."
  ],
  "schema_version": "1.2"
}
