{ "artifacts": [ { "file": "model.litertlm", "sha256": "356b2d7778d27d55fbb0d2e0c5e8801e9de136db59849c41c882bd97333840d5", "size_mb": 2359.731 } ], "benchmarks": [], "conversion": { "command": "from litert_torch.generative.export_hf.export import export; export(model='Qwen/Qwen3-4B-Thinking-2507', output_dir='out', quantization_recipe='qwen3_int4_block128_octav.json', cache_length=4096, externalize_embedder=True) (HF card Conversion snippet)", "quantization": "int4 blockwise-128 + OCTAV, symmetric; embeddings INT8 (externalized); KV cache 4096 (HF card Conversion)", "tool": "litert-torch export_hf (official converter; standard Qwen3ForCausalLM on the existing Qwen3 path, no custom graph code) (HF card Conversion)", "tool_version": "TODO (the HF card names the converter but not its version; no export log in the sources)" }, "cross_runtime": [], "delegation": null, "device": { "records": [ { "device": "galaxy-s26", "run": { "accelerator": "cpu", "context_length": null, "date": "2026-09-05", "decode_tokens_per_s": 5.99, "delegated_ops": 2448, "env": { "device": "Galaxy S26 (SM-S942Q)", "machine_label": "galaxy-s26-cold-cache-cooled", "os_build": "Android 16", "runtime": "litert-lm", "runtime_version": "0.16.0", "soc": "Qualcomm SM8850", "vendor_sdk": null }, "error": null, "evidence": [ "VERBOSE: Replacing 1454 out of 1674 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 285 partitions for subgraph 0 (prefill_128).", "VERBOSE: Replacing 1374 out of 1596 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 290 partitions for subgraph 1 (decode).", "VERBOSE: Replacing 7 out of 7 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 1 partitions for subgraph 2 (odml.rms_norm.impl).", "VERBOSE: Replacing 7 out of 7 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 1 partitions for subgraph 3 (odml.rms_norm.impl_0).", "results block: prefill=63.02 decode=5.99 tokens/s" ], "failure_class": null, "full_delegation": false, "latency_p50_ms": null, "loads": true, "max_abs_diff": null, "max_rel_diff": null, "metrics": { "decode_tokens": 1694.0, "init_s": 3.40004, "prefill_tokens": 204.0 }, "output_match": null, "peak_mem_mb": null, "prefill_tokens_per_s": 63.02, "provenance": "measured", "runs": true, "total_ops": 2526, "ttft_ms": 3400.0 }, "source": "data/device_runs/0.16.0/2026-09-05/qwen3-4b-thinking__galaxy-s26.json" }, { "device": "mac-studio-m4-max", "run": { "accelerator": "gpu", "context_length": null, "date": "2026-07-23", "decode_tokens_per_s": 68.29, "delegated_ops": null, "env": { "device": "Mac Studio (M4 Max)", "machine_label": "mac-studio-m4-max", "os_build": null, "runtime": "litert-lm", "runtime_version": "0.14.0", "soc": "Apple M4 Max", "vendor_sdk": null }, "error": null, "evidence": [ "results block: prefill=970.01 decode=68.29 tokens/s, init=2.0336 s" ], "failure_class": null, "full_delegation": null, "latency_p50_ms": null, "loads": true, "max_abs_diff": null, "max_rel_diff": null, "metrics": { "decode_tokens": 256.0, "init_s": 2.0336, "max_num_tokens": 1024.0, "prefill_tokens": 256.0 }, "output_match": null, "peak_mem_mb": null, "prefill_tokens_per_s": 970.01, "provenance": "measured", "runs": true, "total_ops": null, "ttft_ms": 278.6 }, "source": "data/device_runs/0.14.0/2026-07-23/qwen3-4b-thinking__mac-studio-m4-max.json" }, { "device": "raspberry-pi-5", "run": { "accelerator": "cpu", "context_length": null, "date": "2026-09-01", "decode_tokens_per_s": 1.49, "delegated_ops": null, "env": { "device": "Raspberry Pi 5 Model B Rev 1.1", "machine_label": "raspberry-pi-5-cooled-52c", "os_build": "Linux-6.18.34+rpt-rpi-2712-aarch64-with-glibc2.41", "runtime": "litert-lm", "runtime_version": "0.16.1", "soc": "Broadcom BCM2712", "vendor_sdk": null }, "error": null, "evidence": [ "pi5 LLM sweep row: `litert-lm benchmark --backend cpu --cpu-thread-count 4 -p 256 -d 256 --runs 1 --cache memory` (wave-2 driver pi5_llm_bench.py; --cache memory rather than the house --cache no, which OOM-kills every >=1.2B file on the 8 GB Pi — equivalence measured on granite-350m int8, +2-3%), 3 invocations per file with cool-down to <=52 C between them, vcgencmd measure_temp + get_throttled logged per invocation, peak RSS polled from /proc; throughput = median of the three invocations (spread in metrics); a row counts as measured only when the real-generation gate (`litert-lm run`, degenerate-output check) passed and every invocation exited 0 with get_throttled 0x0", "versions: cpu=Raspberry Pi 5 Model B Rev 1.1, litert-lm=0.16.1, litert-lm-api=0.16.1, platform=Linux-6.18.34+rpt-rpi-2712-aarch64-with-glibc2.41, python=3.13.5", "cache mode 'memory'; -p 256 -d 256 --runs 1 --cpu-thread-count 4", "gate ('What is 17 plus 26? Answer with the number only.'): status pass, exit 0, wall 214.5 s, output head '[thought] Okay, the user is asking \"What is 17 plus 26? Answer with the number o'", "invocation 0: exit=0 wall_s=498.3 temp 51.0->54.9C throttled=0x0 prefill_tps=10.71 decode_tps=1.49 ttft_s=25.2334 init_s=55.9693 peak_rss_mb=3957", "invocation 1: exit=0 wall_s=498.4 temp 49.9->53.2C throttled=0x0 prefill_tps=10.49 decode_tps=1.49 ttft_s=25.6541 init_s=55.8938 peak_rss_mb=3942", "invocation 2: exit=0 wall_s=496.3 temp 51.0->54.3C throttled=0x0 prefill_tps=10.61 decode_tps=1.5 ttft_s=25.4863 init_s=55.4874 peak_rss_mb=3974" ], "failure_class": null, "full_delegation": null, "latency_p50_ms": null, "loads": true, "max_abs_diff": null, "max_rel_diff": null, "metrics": { "decode_tokens": 256.0, "decode_tps_max": 1.5, "decode_tps_min": 1.49, "init_s": 55.8938, "init_s_max": 55.9693, "init_s_min": 55.4874, "invocations": 3.0, "prefill_tokens": 256.0, "prefill_tps_max": 10.71, "prefill_tps_min": 10.49, "runs_per_invocation": 1.0, "threads": 4.0, "ttft_s_max": 25.6541, "ttft_s_min": 25.2334 }, "output_match": null, "peak_mem_mb": 3974.0, "prefill_tokens_per_s": 10.61, "provenance": "measured", "runs": true, "total_ops": null, "ttft_ms": 25486.3 }, "source": "data/device_runs/0.16.1/2026-09-01/qwen3-4b-thinking__raspberry-pi-5.json" } ] }, "model": { "family": "qwen3", "id": "qwen3-4b-thinking", "license": "apache-2.0", "source_url": "https://huggingface.co/litert-community/Qwen3-4B-Thinking-2507", "task": "text-generation" }, "pitfalls": [ "Reasoning model (thinking on by default in the 2507 release); this repo's rows are the Mac compat/gpu_audit runs, the S26 CPU S7 row and the Pi 5 CPU row (device runs).", "License Apache-2.0, inherited from Qwen/Qwen3-4B-Thinking-2507 (HF card 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/ (Pi 5 / S26 / Pixel 8a / Mac / iPhone as measured)." ], "schema_version": "1.2" }