laya-LiteRT / hfmodels.json
mlboydaisuke's picture
hfmodels.json: npu profiles for ml_s256_wfp16 and en_s256_fp32 (Galaxy S26 PASS records, LiteRT 2.2.0); the gpu records of ml_s256_fp32 and en_s256_fp32 re-run on the 2026-09-29 graphs
f557459 verified
Raw History Blame Contribute Delete
23.4 kB
{
"schema_version": 1,
"model_id": "litert-community/laya-LiteRT",
"tasks": [
"decide"
],
"default_variant": "ml_s256_fp32",
"license": {
"id": "apache-2.0",
"url": "https://huggingface.co/litert-community/laya-LiteRT/blob/main/LICENSE"
},
"variants": [
{
"id": "ml_s256_fp32",
"runtime": "litert",
"handler": {
"id": "litert.typed_decisions",
"abi": 1
},
"runtime_range": {
"min_inclusive": "2.2.0",
"max_exclusive": "3.0.0"
},
"inputs": [
"text"
],
"files": [
{
"id": "main",
"role": "model",
"path": "laya_ml_s256_fp32.tflite",
"bytes": 1287376352,
"sha256": "98ba7a57d054ad96c80155283f40a5034eda7d5c14bce60115436c97f4595c2a"
},
{
"id": "act",
"role": "model",
"path": "laya_ml_act_head_fp32.tflite",
"bytes": 795816,
"sha256": "30da532a8b752fdc82351740b0194d440a13d342ed77b9f8848978f1a96cf16d"
},
{
"id": "tokenizer",
"role": "tokenizer",
"path": "multilingual/tokenizer.json",
"bytes": 34363188,
"sha256": "609d8f4c067cd3950f88594c5a802616cea245823836ef5848ee4fc40aab5b6f"
},
{
"id": "tokenizer_config",
"role": "tokenizer",
"path": "multilingual/tokenizer_config.json",
"bytes": 524,
"sha256": "6c6b2d8e3c84ce0e671c129cd6b374b235d6f9863042a5836358d00a89bbb5a1"
},
{
"id": "config",
"role": "processor_config",
"path": "multilingual/calibration.json",
"bytes": 9156,
"sha256": "80e148c68154b607e7bb66446f064b1fe5da649e1074c917b25e02551a566f2c"
}
],
"default_profile": "gpu",
"profiles": [
{
"id": "gpu",
"priority": 100,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "gpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [
"cpu"
],
"default_selectable": true,
"verification": [
{
"level": "PUBLISHER_TESTED",
"result": "PASS",
"note": "graph bytes of 2026-09-29 (fp16-safe rewrite, fp32-exact): 201/201 rows argmax identical, max |dp| 0.0001; 55.7 ms per question (median; p90 124.7 as the phone reached thermal status 1-2); 1713 of 1713 nodes on LITERT_CL; litert/results/2026-09-29-2351-RFGL80R6A6H-litert2.2.0-decide-ml_s256_fp32-gpu.log",
"device": "SM-S942Q (Galaxy S26)",
"os_build": "BP4A.251205.006.S942QOPS1AZF2",
"runtime": "litert 2.2.0",
"date": "2026-09-29"
}
]
},
{
"id": "cpu",
"priority": 50,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "cpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [],
"default_selectable": true,
"verification": [
{
"level": "PUBLISHER_TESTED",
"result": "PASS",
"note": "30/30 rows; 222 ms per question; litert/results/2026-09-21-1421-RFGL80R6A6H-litert2.2.0-decide-ml_s256_fp32-cpu.log",
"device": "SM-S942Q (Galaxy S26)",
"os_build": "BP4A.251205.006.S942QOPS1AZF2",
"runtime": "litert 2.2.0",
"date": "2026-09-21"
}
]
}
],
"handler_config": {
"family": "laya",
"window": 256,
"hidden": 768,
"head_tokens": 256,
"files": {
"main": "main",
"act_head": "act",
"tokenizer": "tokenizer",
"tokenizer_config": "tokenizer_config",
"config": "config"
},
"gpu_precision": "fp32",
"cpu_threads": 4,
"languages": [
"multilingual"
]
}
},
{
"id": "ml_s512_fp32",
"runtime": "litert",
"handler": {
"id": "litert.typed_decisions",
"abi": 1
},
"runtime_range": {
"min_inclusive": "2.2.0",
"max_exclusive": "3.0.0"
},
"inputs": [
"text"
],
"files": [
{
"id": "main",
"role": "model",
"path": "laya_ml_s512_fp32.tflite",
"bytes": 1288293856,
"sha256": "6fa52d9c19f129c2f3e8db8dcc60a597f7919c3bd56c35a41b947aa3188dc58b"
},
{
"id": "act",
"role": "model",
"path": "laya_ml_act_head_fp32.tflite",
"bytes": 795816,
"sha256": "30da532a8b752fdc82351740b0194d440a13d342ed77b9f8848978f1a96cf16d"
},
{
"id": "tokenizer",
"role": "tokenizer",
"path": "multilingual/tokenizer.json",
"bytes": 34363188,
"sha256": "609d8f4c067cd3950f88594c5a802616cea245823836ef5848ee4fc40aab5b6f"
},
{
"id": "tokenizer_config",
"role": "tokenizer",
"path": "multilingual/tokenizer_config.json",
"bytes": 524,
"sha256": "6c6b2d8e3c84ce0e671c129cd6b374b235d6f9863042a5836358d00a89bbb5a1"
},
{
"id": "config",
"role": "processor_config",
"path": "multilingual/calibration.json",
"bytes": 9156,
"sha256": "80e148c68154b607e7bb66446f064b1fe5da649e1074c917b25e02551a566f2c"
}
],
"default_profile": "gpu",
"profiles": [
{
"id": "gpu",
"priority": 100,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "gpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [
"cpu"
],
"default_selectable": true
},
{
"id": "cpu",
"priority": 50,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "cpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [],
"default_selectable": true
}
],
"handler_config": {
"family": "laya",
"window": 512,
"hidden": 768,
"head_tokens": 256,
"files": {
"main": "main",
"act_head": "act",
"tokenizer": "tokenizer",
"tokenizer_config": "tokenizer_config",
"config": "config"
},
"gpu_precision": "fp32",
"cpu_threads": 4,
"languages": [
"multilingual"
]
}
},
{
"id": "ml_s256_wfp16",
"runtime": "litert",
"handler": {
"id": "litert.typed_decisions",
"abi": 1
},
"runtime_range": {
"min_inclusive": "2.2.0",
"max_exclusive": "3.0.0"
},
"inputs": [
"text"
],
"files": [
{
"id": "main",
"role": "model",
"path": "laya_ml_s256_wfp16.tflite",
"bytes": 644077360,
"sha256": "84016a50a42f915a06567f2c31f95a95cc47c87ae1100c9fe60e699f6272b1d2"
},
{
"id": "act",
"role": "model",
"path": "laya_ml_act_head_fp32.tflite",
"bytes": 795816,
"sha256": "30da532a8b752fdc82351740b0194d440a13d342ed77b9f8848978f1a96cf16d"
},
{
"id": "tokenizer",
"role": "tokenizer",
"path": "multilingual/tokenizer.json",
"bytes": 34363188,
"sha256": "609d8f4c067cd3950f88594c5a802616cea245823836ef5848ee4fc40aab5b6f"
},
{
"id": "tokenizer_config",
"role": "tokenizer",
"path": "multilingual/tokenizer_config.json",
"bytes": 524,
"sha256": "6c6b2d8e3c84ce0e671c129cd6b374b235d6f9863042a5836358d00a89bbb5a1"
},
{
"id": "config",
"role": "processor_config",
"path": "multilingual/calibration.json",
"bytes": 9156,
"sha256": "80e148c68154b607e7bb66446f064b1fe5da649e1074c917b25e02551a566f2c"
}
],
"default_profile": "cpu",
"profiles": [
{
"id": "npu",
"priority": 120,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "npu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [],
"default_selectable": true,
"verification": [
{
"level": "PUBLISHER_TESTED",
"result": "PASS",
"note": "201/201 captured rows argmax identical, max |dp| 0.0068; 32.6 ms per question (median, warm; 33.9 ms in a second process); whole graph on the NPU (1813 of 1813 ops, 1 DispatchDelegate node); first compile 76.0 s, 0.93 s from LiteRT's cache; litert/results/2026-09-29-2348-RFGL80R6A6H-litert2.2.0-decide-ml_s256_wfp16-npu.log, 2026-09-29-2349-...-npu-cached.log",
"device": "SM-S942Q (Galaxy S26)",
"os_build": "BP4A.251205.006.S942QOPS1AZF2",
"runtime": "litert 2.2.0",
"date": "2026-09-29"
}
]
},
{
"id": "cpu",
"priority": 100,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "cpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [],
"default_selectable": true
}
],
"handler_config": {
"family": "laya",
"window": 256,
"hidden": 768,
"head_tokens": 256,
"files": {
"main": "main",
"act_head": "act",
"tokenizer": "tokenizer",
"tokenizer_config": "tokenizer_config",
"config": "config"
},
"gpu_precision": "fp32",
"cpu_threads": 4,
"languages": [
"multilingual"
]
}
},
{
"id": "en_s256_fp32",
"runtime": "litert",
"handler": {
"id": "litert.typed_decisions",
"abi": 1
},
"runtime_range": {
"min_inclusive": "2.2.0",
"max_exclusive": "3.0.0"
},
"inputs": [
"text"
],
"files": [
{
"id": "main",
"role": "model",
"path": "laya_en_s256_fp32.tflite",
"bytes": 1684756736,
"sha256": "683c03749b123fdb7ec168f257b9d54462c3fc8205539cb2fb83bc617962ecc8"
},
{
"id": "act",
"role": "model",
"path": "laya_act_head_fp32.tflite",
"bytes": 1057960,
"sha256": "c6f8de9b66e36581ce03cca7047d0a6b8573ba448ce44ae613e49c58f34e4052"
},
{
"id": "tokenizer",
"role": "tokenizer",
"path": "en/tokenizer.json",
"bytes": 3583228,
"sha256": "6c8aaa9a542084f2457eab775d4eeb51f92a70c0fd9de28d5edb0ddec3c08d30"
},
{
"id": "tokenizer_config",
"role": "tokenizer",
"path": "en/tokenizer_config.json",
"bytes": 308,
"sha256": "50044de60daaa73df97d262e15a40d4faf0160e7d742df64b377877a1320dd12"
},
{
"id": "config",
"role": "processor_config",
"path": "en/rl_agent_config.json",
"bytes": 745,
"sha256": "ae287b56bbcf5f8c4f4541ae9dfd00c914c4c48b940b8398c3058af37ba92bbd"
}
],
"default_profile": "gpu",
"profiles": [
{
"id": "npu",
"priority": 120,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "npu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [],
"default_selectable": true,
"verification": [
{
"level": "PUBLISHER_TESTED",
"result": "PASS",
"note": "140/140 rows argmax identical, max |dp| 0.0054; 65.3 ms per question (median, warm); whole graph on the NPU (2159 of 2159 ops, 1 DispatchDelegate node); first compile 65.2 s; litert/results/2026-09-29-2349-RFGL80R6A6H-litert2.2.0-decide-en_s256_fp32-npu.log",
"device": "SM-S942Q (Galaxy S26)",
"os_build": "BP4A.251205.006.S942QOPS1AZF2",
"runtime": "litert 2.2.0",
"date": "2026-09-29"
}
]
},
{
"id": "gpu",
"priority": 100,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "gpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [
"cpu"
],
"default_selectable": true,
"verification": [
{
"level": "PUBLISHER_TESTED",
"result": "PASS",
"note": "graph bytes of 2026-09-29: 140/140 rows argmax identical, max |dp| 0.0000, 140/140 rounded answers identical; 234.5 ms per question at thermal status 2 (127 ms on 2026-09-21 on the earlier bytes); 2159 of 2159 nodes on LITERT_CL; litert/results/2026-09-29-2352-RFGL80R6A6H-litert2.2.0-decide-en_s256_fp32-gpu.log",
"device": "SM-S942Q (Galaxy S26)",
"os_build": "BP4A.251205.006.S942QOPS1AZF2",
"runtime": "litert 2.2.0",
"date": "2026-09-29"
}
]
},
{
"id": "cpu",
"priority": 50,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "cpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [],
"default_selectable": true
}
],
"handler_config": {
"family": "laya",
"window": 256,
"hidden": 1024,
"head_tokens": 192,
"files": {
"main": "main",
"act_head": "act",
"tokenizer": "tokenizer",
"tokenizer_config": "tokenizer_config",
"config": "config"
},
"gpu_precision": "fp32",
"cpu_threads": 4,
"languages": [
"en"
]
}
},
{
"id": "en_s512_fp32",
"runtime": "litert",
"handler": {
"id": "litert.typed_decisions",
"abi": 1
},
"runtime_range": {
"min_inclusive": "2.2.0",
"max_exclusive": "3.0.0"
},
"inputs": [
"text"
],
"files": [
{
"id": "main",
"role": "model",
"path": "laya_en_s512_fp32.tflite",
"bytes": 1685805312,
"sha256": "6bcd53c001540d63e3366df80ebc850ab1623f887445f8fae7993f6ae65bb10a"
},
{
"id": "act",
"role": "model",
"path": "laya_act_head_fp32.tflite",
"bytes": 1057960,
"sha256": "c6f8de9b66e36581ce03cca7047d0a6b8573ba448ce44ae613e49c58f34e4052"
},
{
"id": "tokenizer",
"role": "tokenizer",
"path": "en/tokenizer.json",
"bytes": 3583228,
"sha256": "6c8aaa9a542084f2457eab775d4eeb51f92a70c0fd9de28d5edb0ddec3c08d30"
},
{
"id": "tokenizer_config",
"role": "tokenizer",
"path": "en/tokenizer_config.json",
"bytes": 308,
"sha256": "50044de60daaa73df97d262e15a40d4faf0160e7d742df64b377877a1320dd12"
},
{
"id": "config",
"role": "processor_config",
"path": "en/rl_agent_config.json",
"bytes": 745,
"sha256": "ae287b56bbcf5f8c4f4541ae9dfd00c914c4c48b940b8398c3058af37ba92bbd"
}
],
"default_profile": "gpu",
"profiles": [
{
"id": "gpu",
"priority": 100,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "gpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [
"cpu"
],
"default_selectable": true,
"verification": [
{
"level": "PUBLISHER_TESTED",
"result": "PASS",
"note": "209/209 rows; 442 ms per question; litert/results/2026-09-21-1443-RFGL80R6A6H-litert2.2.0-decide-en_s512_fp32-gpu.log",
"device": "SM-S942Q (Galaxy S26)",
"os_build": "BP4A.251205.006.S942QOPS1AZF2",
"runtime": "litert 2.2.0",
"date": "2026-09-21"
}
]
},
{
"id": "cpu",
"priority": 50,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "cpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [],
"default_selectable": true
}
],
"handler_config": {
"family": "laya",
"window": 512,
"hidden": 1024,
"head_tokens": 192,
"files": {
"main": "main",
"act_head": "act",
"tokenizer": "tokenizer",
"tokenizer_config": "tokenizer_config",
"config": "config"
},
"gpu_precision": "fp32",
"cpu_threads": 4,
"languages": [
"en"
]
}
},
{
"id": "en_s512_wfp16",
"runtime": "litert",
"handler": {
"id": "litert.typed_decisions",
"abi": 1
},
"runtime_range": {
"min_inclusive": "2.2.0",
"max_exclusive": "3.0.0"
},
"inputs": [
"text"
],
"files": [
{
"id": "main",
"role": "model",
"path": "laya_en_s512_wfp16.tflite",
"bytes": 843929776,
"sha256": "6be0cb36ec29c94e94940ee2eaf44bebdb7e6f995c585fbc20471132f6eb8f90"
},
{
"id": "act",
"role": "model",
"path": "laya_act_head_fp32.tflite",
"bytes": 1057960,
"sha256": "c6f8de9b66e36581ce03cca7047d0a6b8573ba448ce44ae613e49c58f34e4052"
},
{
"id": "tokenizer",
"role": "tokenizer",
"path": "en/tokenizer.json",
"bytes": 3583228,
"sha256": "6c8aaa9a542084f2457eab775d4eeb51f92a70c0fd9de28d5edb0ddec3c08d30"
},
{
"id": "tokenizer_config",
"role": "tokenizer",
"path": "en/tokenizer_config.json",
"bytes": 308,
"sha256": "50044de60daaa73df97d262e15a40d4faf0160e7d742df64b377877a1320dd12"
},
{
"id": "config",
"role": "processor_config",
"path": "en/rl_agent_config.json",
"bytes": 745,
"sha256": "ae287b56bbcf5f8c4f4541ae9dfd00c914c4c48b940b8398c3058af37ba92bbd"
}
],
"default_profile": "cpu",
"profiles": [
{
"id": "cpu",
"priority": 100,
"files": [
"main",
"act",
"tokenizer",
"tokenizer_config",
"config"
],
"enabled_inputs": [
"text"
],
"components": {
"inference": "cpu"
},
"requirements": {
"min_android_api": 31,
"abis": [
"arm64-v8a"
]
},
"fallback_profiles": [],
"default_selectable": true,
"verification": [
{
"level": "PUBLISHER_TESTED",
"result": "PASS",
"note": "30/30 rows, max |dp| 0.0005; 1326 ms per question; litert/results/2026-09-21-1417-RFGL80R6A6H-litert2.2.0-decide-en_s512_wfp16-cpu.log",
"device": "SM-S942Q (Galaxy S26)",
"os_build": "BP4A.251205.006.S942QOPS1AZF2",
"runtime": "litert 2.2.0",
"date": "2026-09-21"
}
]
}
],
"handler_config": {
"family": "laya",
"window": 512,
"hidden": 1024,
"head_tokens": 192,
"files": {
"main": "main",
"act_head": "act",
"tokenizer": "tokenizer",
"tokenizer_config": "tokenizer_config",
"config": "config"
},
"gpu_precision": "fp32",
"cpu_threads": 4,
"languages": [
"en"
]
}
}
]
}