{ "artifacts": [ { "file": "Falcon-H1-1.5B-Deep-Instruct_int8.litertlm", "sha256": "759b3b361ccd9b6138767fe1771f35d3be64301308b9a178d3edd43819447526", "size_mb": 1748.166 } ], "benchmarks": [], "conversion": { "command": "TODO (hf-to-litertlm falcon_h1_work/; invocation not stated in the README)", "quantization": "post-hoc dynamic int8 over linears + embedding only; convs and the scan stay float; fp32 activations declared for GPU (README file table + Conversion notes)", "tool": "litert-torch plus a hybrid-cache patch (reproduction script + patch: hf-to-litertlm falcon_h1_work/) (README Conversion notes)", "tool_version": "TODO (the README names the converter but not its version)" }, "cross_runtime": [], "delegation": null, "device": { "records": [ { "device": "galaxy-s26", "run": { "accelerator": "cpu", "context_length": null, "date": "2026-09-05", "decode_tokens_per_s": 11.97, "delegated_ops": 10698, "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 9519 out of 10189 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 660 partitions for subgraph 0 (prefill_1024).", "VERBOSE: Replacing 9519 out of 10189 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 660 partitions for subgraph 1 (prefill_512).", "VERBOSE: Replacing 9519 out of 10189 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 660 partitions for subgraph 2 (prefill_256).", "VERBOSE: Replacing 9453 out of 10123 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 660 partitions for subgraph 3 (prefill_128).", "results block: prefill=103.14 decode=11.97 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": 208.0, "init_s": 6.03715, "prefill_tokens": 212.0 }, "output_match": null, "peak_mem_mb": null, "prefill_tokens_per_s": 103.14, "provenance": "measured", "runs": true, "total_ops": 11237, "ttft_ms": 2140.0 }, "source": "data/device_runs/0.16.0/2026-09-05/falcon-h1-1.5b-deep-instruct-int8__galaxy-s26.json" }, { "device": "galaxy-s26", "run": { "accelerator": "gpu", "context_length": null, "date": "2026-08-24", "decode_tokens_per_s": 6.0, "delegated_ops": 10451, "env": { "device": "Galaxy S26 (SM-S942Q)", "machine_label": "galaxy-s26-cl-pinned", "os_build": "Android 16", "runtime": "litert-lm", "runtime_version": "0.16.0", "soc": "Qualcomm SM8850", "vendor_sdk": null }, "error": null, "evidence": [ "VERBOSE: Replacing 10189 out of 10189 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (prefill_1024).", "VERBOSE: Replacing 10189 out of 10189 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 1 (prefill_512).", "VERBOSE: Replacing 10189 out of 10189 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 2 (prefill_256).", "VERBOSE: Replacing 10123 out of 10123 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 3 (prefill_128).", "results block: prefill=120.58 decode=6.0 tokens/s" ], "failure_class": null, "full_delegation": true, "latency_p50_ms": null, "loads": true, "max_abs_diff": null, "max_rel_diff": null, "metrics": { "decode_tokens": 170.0, "init_s": 86.12778, "prefill_tokens": 212.0 }, "output_match": null, "peak_mem_mb": null, "prefill_tokens_per_s": 120.58, "provenance": "measured", "runs": true, "total_ops": 10451, "ttft_ms": 1920.0 }, "source": "data/device_runs/0.16.0/2026-08-24/falcon-h1-1.5b-deep-instruct-int8__galaxy-s26.json" }, { "device": "raspberry-pi-5", "run": { "accelerator": "cpu", "context_length": null, "date": "2026-09-02", "decode_tokens_per_s": 2.05, "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 66.1 s, output head '43'", "invocation 0: exit=0 wall_s=338.2 temp 47.2->53.2C throttled=0x0 prefill_tps=24.59 decode_tps=2.04 ttft_s=10.9016 init_s=65.4514 peak_rss_mb=3367", "invocation 1: exit=0 wall_s=338.2 temp 51.0->55.4C throttled=0x0 prefill_tps=24.73 decode_tps=2.05 ttft_s=10.8384 init_s=64.7474 peak_rss_mb=3367", "invocation 2: exit=0 wall_s=338.2 temp 49.9->53.8C throttled=0x0 prefill_tps=24.59 decode_tps=2.05 ttft_s=10.8977 init_s=64.7764 peak_rss_mb=3367" ], "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": 2.05, "decode_tps_min": 2.04, "init_s": 64.7764, "init_s_max": 65.4514, "init_s_min": 64.7474, "invocations": 3.0, "prefill_tokens": 256.0, "prefill_tps_max": 24.73, "prefill_tps_min": 24.59, "runs_per_invocation": 1.0, "threads": 4.0, "ttft_s_max": 10.9016, "ttft_s_min": 10.8384 }, "output_match": null, "peak_mem_mb": 3367.0, "prefill_tokens_per_s": 24.59, "provenance": "measured", "runs": true, "total_ops": null, "ttft_ms": 10897.7 }, "source": "data/device_runs/0.16.1/2026-09-02/falcon-h1-1.5b-deep-instruct-int8__raspberry-pi-5.json" } ] }, "model": { "family": "falcon-h1", "id": "falcon-h1-1.5b-deep-instruct-int8", "license": "other (falcon-llm-license)", "source_url": "https://huggingface.co/litert-community/Falcon-H1-1.5B-Deep-Instruct", "task": "text-generation" }, "pitfalls": [ "Requires litert-lm >= 0.15 (README intro).", "Correctness (README): float export matches the HF model teacher-forced across 48 decode positions (top-1/top-5 identical, corr 1.0000, KL ≈ 0); 8-question gate 8/8 on every lane (GPU and CPU, litert-lm 0.15.0 and 0.16.0); hermetic prefill-chunk sweep clean (CPU fills 12-51, GPU 12-31); iPhone 17 Pro composite probe 8/8 on GPU and CPU.", "Hybrid packaging: composite hybrid cache layer (KV + conv + recurrent state at one layer index), folded selective scan as batched matmuls with chunk/head axes in the batch axis (all tensors rank <= 4, no BROADCAST_TO, no int64 index math — what makes the graph fully delegable on GPU), Falcon µP multiplier vector and ssm_in_multiplier preserved, prefill-pad guard (README Conversion notes).", "Galaxy S26 GPU per the README: runs, 115393/115393 ops across 12 subgraphs on LiteRT GPU; this repo's S26 CPU row is from the S7 backfill (device runs).", "License: Falcon LLM License inherited from the base model; changes = weights converted and quantized as stated, tokenizer and chat template repackaged unmodified; community conversion (README License and changes).", "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" }