{ "artifacts": [ { "file": "ram_stage3_tail_fp16.tflite", "sha256": "c46e40cab69f070a4b7ba86b9de29f15148f9c67804c5e0888290d243f24ed4e", "size_mb": 117.326 } ], "benchmarks": [], "conversion": { "command": "python build_hybrid.py", "quantization": "fp16", "tool": "litert-torch (ramplus-work build scripts, litert_torch.convert; fp16 via ai_edge_quantizer float_casting)", "tool_version": "0.10.0 (editable dev checkout 115a136 + local patches)" }, "cross_runtime": [], "delegation": null, "device": { "records": [ { "device": "galaxy-s26", "run": { "accelerator": "gpu_mldrift", "context_length": null, "date": "2026-08-26", "decode_tokens_per_s": null, "delegated_ops": null, "env": { "device": "Galaxy S26 (SM-S942Q)", "machine_label": "galaxy-s26-npubench", "os_build": "Android 16", "runtime": "litert", "runtime_version": "2.2.0", "soc": "Qualcomm SM8850", "vendor_sdk": null }, "error": null, "evidence": [ "npubench sweep row: N=50 median, 1 backend = 1 process, accepted only with thermal NONE->NONE; NPU rows additionally required qnn_partition delegate evidence in logcat (a stock model asked for on the NPU can silently land on XNNPACK and report a CPU number)", "mode=jit thermal=NONE->NONE headroom=0.77666664->0.7733334 attempt=0" ], "failure_class": null, "full_delegation": null, "latency_p50_ms": 8.005, "loads": true, "max_abs_diff": null, "max_rel_diff": null, "metrics": { "iterations": 50, "latency_max_ms": 9.075, "latency_min_ms": 7.592, "load_ms": 3334.8 }, "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/2.2.0/2026-08-26/ram-plus__ram_stage3_tail_fp16__galaxy-s26.json" }, { "device": "galaxy-s26", "run": { "accelerator": "npu_qnn", "context_length": null, "date": "2026-08-26", "decode_tokens_per_s": null, "delegated_ops": null, "env": { "device": "Galaxy S26 (SM-S942Q)", "machine_label": "galaxy-s26-npubench", "os_build": "Android 16", "runtime": "litert", "runtime_version": "2.2.0", "soc": "Qualcomm SM8850", "vendor_sdk": "QAIRT (Hexagon, JIT on-device)" }, "error": null, "evidence": [ "npubench sweep row: N=50 median, 1 backend = 1 process, accepted only with thermal NONE->NONE; NPU rows additionally required qnn_partition delegate evidence in logcat (a stock model asked for on the NPU can silently land on XNNPACK and report a CPU number)", "mode=jit thermal=NONE->NONE headroom=0.78000003->0.77666664 attempt=0" ], "failure_class": null, "full_delegation": null, "latency_p50_ms": 14.568, "loads": true, "max_abs_diff": null, "max_rel_diff": null, "metrics": { "iterations": 50, "jit_first_load_ms": 7336.6, "latency_max_ms": 14.683, "latency_min_ms": 14.403, "load_ms": 147.9, "qnn_partition_lines": 57 }, "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/2.2.0/2026-08-26/ram-plus__ram_stage3_tail_fp16__galaxy-s26.json" }, { "device": "raspberry-pi-5", "run": { "accelerator": "cpu_xnnpack", "context_length": null, "date": "2026-08-31", "decode_tokens_per_s": null, "delegated_ops": null, "env": { "device": "Raspberry Pi 5 Model B Rev 1.1", "machine_label": "raspberry-pi-5", "os_build": "Linux-6.18.34+rpt-rpi-2712-aarch64-with-glibc2.41", "runtime": "litert", "runtime_version": "2.2.0.dev20260804", "soc": null, "vendor_sdk": null }, "error": null, "evidence": [ "pi5 sweep row: LiteRT benchmark_model, CPU/XNNPACK at --num_threads=4, 3 invocations per file of 10 warm-up + 50 timed runs (the tool caps a phase at 150 s, so very slow graphs run fewer); latency = median of the three per-invocation medians over the timed phase; a row counts as measured only when every invocation exited 0 with XNNPACK engaged and vcgencmd get_throttled 0x0 before and after", "versions: ai-edge-litert-nightly=2.2.0.dev20260804, benchmark_model_sha256=babc9275addd8612caf0842294cb5b3f301c3cce46e6c5979cd2803f215f6bee, cpu=Raspberry Pi 5 Model B Rev 1.1, litert-cli-nightly=0.2.0.dev20260805, platform=Linux-6.18.34+rpt-rpi-2712-aarch64-with-glibc2.41, python=3.13.5", "invocation 0: exit=0 wall_s=14.3 temp 49.4->63.7C throttled=0x0 xnnpack=True median_us=235099.0 runs=50 footprint_peak_mb=390.23", "invocation 1: exit=0 wall_s=14.0 temp 51.0->63.7C throttled=0x0 xnnpack=True median_us=229520.0 runs=50 footprint_peak_mb=390.73", "invocation 2: exit=0 wall_s=14.6 temp 50.5->64.2C throttled=0x0 xnnpack=True median_us=240265.0 runs=50 footprint_peak_mb=390.53" ], "failure_class": null, "full_delegation": null, "latency_p50_ms": 235.099, "loads": true, "max_abs_diff": null, "max_rel_diff": null, "metrics": { "init_ms": 152.874, "invocations": 3, "iterations": 150, "latency_max_ms": 246.768, "latency_min_ms": 224.436, "threads": 4, "warmup_runs_per_invocation": 10 }, "output_match": null, "peak_mem_mb": 390.73, "prefill_tokens_per_s": null, "provenance": "measured", "runs": true, "total_ops": null, "ttft_ms": null }, "source": "data/device_runs/2.2.0.dev20260804/2026-08-31/ram-plus__ram_stage3_tail_fp16__raspberry-pi-5.json" } ] }, "model": { "family": "ram-plus", "id": "ram-plus__ram_stage3_tail_fp16", "license": "apache-2.0", "source_url": "https://huggingface.co/litert-community/RAM-Plus-LiteRT", "task": "image-classification" }, "pitfalls": [ "Swin-L stage 3 must run on CPU: it fp16-miscomputes on the Mali GPU delegate (fp16 matmul accumulation in the deep, high-magnitude blocks — not overflow, not head_dim) (ram/README.md L33-41; HF card)." ], "schema_version": "1.2" }