duplex-voice-models / catalog.json
ELAINAy's picture
Add files using upload-large-folder tool
f298324 verified
Raw History Blame Contribute Delete
14.1 kB
{
"schema": 1,
"revision": 5,
"models": [
{
"id": "minicpmo-4.5",
"engine": "minicpmo",
"recommended": true,
"name": "MiniCPM-o 4.5",
"publisher": "OpenBMB",
"summary": "9B 全模态模型的全双工语音部分(中英文):Whisper-medium 听、Qwen3-8B 每秒决定听或说、TTS + CosyVoice2 流式发声。",
"languages": [
"zh",
"en"
],
"license": "Apache-2.0",
"upstream": "openbmb/MiniCPM-o-4_5",
"variants": [
{
"id": "minicpmo45-v1-sm8850",
"title": "LLM LPBQ W4 · Whisper / DiT / HiFT W8 · TTS CPU int8",
"path": "minicpmo/minicpmo45-v1-sm8850",
"socs": [
"SM8850"
],
"htp": "v81",
"qairt": "2.45.0.260326",
"notes": "QNN HTP context 绑定 SoC 与 QAIRT 版本;1 秒一个单元,TTS 在 CPU 上;实测说话时 HTP 约 810 ms/s,能跟上实时,语音比文字晚约 2 秒(含 1 秒播放预留)。",
"files": [
{
"path": "assets.bin",
"size": 1733394180,
"sha256": "48e86df743f1b0b9d388c3de9ae57de1c5ad35587250be03699bc5b0d636a06e"
},
{
"path": "audio_encoder.bin",
"size": 338874368,
"sha256": "861ad7da7799a1af457692778bb10987d2554888a758ceb3ca0e7a3cec817a79"
},
{
"path": "dit.bin",
"size": 80216064,
"sha256": "b2c12369568a11e843db7a8fdbd5dfadd602fe8409e889fdd636a8396d028724"
},
{
"path": "flow_encoder.onnx",
"size": 185649479,
"sha256": "6cf1a240eb98dfd1122c79e4c1ebd368fca84ebecad74a07029d12251f33efe9"
},
{
"path": "hift_decode.bin",
"size": 24342528,
"sha256": "e01bba7007048dea84aac2aeafe7ed84a124cc83e7cc1bd230f43095660afe94"
},
{
"path": "hift_f0.bin",
"size": 3416064,
"sha256": "765d3f5c2af4887bd2693bc2476cabe5c72ce464d4724e21cb2808fda529e6fe"
},
{
"path": "llm_head.bin",
"size": 628146176,
"sha256": "24d13026032ed7bc41fdd3714e9d903086482d5a4a97fa6159fae9dd71d048a4"
},
{
"path": "llm_part_0.bin",
"size": 913174528,
"sha256": "7700e58c3cc3b11085f6607c0b5395ece88456426d71f56363fac31e73a121b6"
},
{
"path": "llm_part_1.bin",
"size": 913268736,
"sha256": "8dff714578d79c4cf0f534d26d98b20786d7db2ea8a5dcce3fdf07566e10ef60"
},
{
"path": "llm_part_2.bin",
"size": 913170432,
"sha256": "3c8933b81955b45b526868a22424e1a14914859e150cd25c9196f11256f9655b"
},
{
"path": "llm_part_3.bin",
"size": 913256448,
"sha256": "1f9b08be92659cadfcd53819989a9e5a283503de9c2ce4f2ca0b941e97e76bfa"
},
{
"path": "tts_cpu.bin",
"size": 194742088,
"sha256": "74969808183cc9f91b37d0eee28c617a817b302af3ff445b98cbdea0e4301676"
}
]
},
{
"id": "minicpmo45-v1-sm8975",
"title": "LLM LPBQ W4 · Whisper / DiT / HiFT W8 · TTS CPU int8",
"path": "minicpmo/minicpmo45-v1-sm8975",
"socs": [
"SM8975"
],
"htp": "v85",
"qairt": "2.51.40.260930",
"notes": "与 SM8850 版同一套量化,用 QAIRT 2.51 为 SM8975(HTP v85)重新编译 context;Android 17 上需要带 NPU Manager 授权的新版 App。实测加载 5.6 秒,19 秒演示通话平均每 80 ms 帧 31 ms(RTF 0.39)。",
"files": [
{
"path": "assets.bin",
"size": 1733394180,
"sha256": "48e86df743f1b0b9d388c3de9ae57de1c5ad35587250be03699bc5b0d636a06e"
},
{
"path": "audio_encoder.bin",
"size": 335949824,
"sha256": "aaa57d663146e9b335b1a63508131caafd7a0fb86bf8eb405a89305977c51b9f"
},
{
"path": "dit.bin",
"size": 78688256,
"sha256": "d8e3c9c1347ee419712f10a65e49bdc87113b30d3cc0dec5d61c2d61ba9cebd6"
},
{
"path": "flow_encoder.onnx",
"size": 185649479,
"sha256": "6cf1a240eb98dfd1122c79e4c1ebd368fca84ebecad74a07029d12251f33efe9"
},
{
"path": "hift_decode.bin",
"size": 21569536,
"sha256": "c6837f86960ce01a1f151a8bf4723fd3a95c91c43ce7e7a7f237c8a12c2bb022"
},
{
"path": "hift_f0.bin",
"size": 3399680,
"sha256": "1fd234aae7cee68c1c4e3db8f9460f61b2d695f04dcb2a8478ba37a30a26f33a"
},
{
"path": "llm_head.bin",
"size": 626921472,
"sha256": "77588d705df56cb525f329149a1a97ac5c96a635b6338fb93c492f1461939328"
},
{
"path": "llm_part_0.bin",
"size": 926277632,
"sha256": "f4c34f35b6c7ab50dab27132486d90033f267915f4b2675c26d4593f69f026a0"
},
{
"path": "llm_part_1.bin",
"size": 926265344,
"sha256": "cb0642be52cfe18736d66807dc01c210acfba202988caf2b3fea1389c35652a6"
},
{
"path": "llm_part_2.bin",
"size": 926277632,
"sha256": "851d83fdd0c206176677ade08189b01bdfb4436525a04ebcbf010ab228156d28"
},
{
"path": "llm_part_3.bin",
"size": 926281728,
"sha256": "3eeb48f998a50f6de769bb4933977383aadf2f491d76bc3cb554aeb6e556c927"
},
{
"path": "tts_cpu.bin",
"size": 194742088,
"sha256": "74969808183cc9f91b37d0eee28c617a817b302af3ff445b98cbdea0e4301676"
}
]
},
{
"id": "minicpmo45-v2-sm8975",
"title": "LLM LPBQ W4 · Whisper / SigLIP / DiT / HiFT W8 · TTS CPU int8 · 可看图",
"path": "minicpmo/minicpmo45-v2-sm8975",
"socs": [
"SM8975"
],
"htp": "v85",
"qairt": "2.51.40.260930",
"notes": "v1 加上视觉(SigLIP + resampler,W8A16):通话前或通话中发一张照片,模型能看图回答。与 SM8850 版同一套量化,用 QAIRT 2.51 为 SM8975(HTP v85)重新编译 context;Android 17 上需要带 NPU Manager 授权的新版 App。实测加载 5.6 秒,19 秒演示通话平均每 80 ms 帧 31 ms(RTF 0.39)。",
"files": [
{
"path": "assets.bin",
"size": 1733394180,
"sha256": "48e86df743f1b0b9d388c3de9ae57de1c5ad35587250be03699bc5b0d636a06e"
},
{
"path": "audio_encoder.bin",
"size": 335949824,
"sha256": "aaa57d663146e9b335b1a63508131caafd7a0fb86bf8eb405a89305977c51b9f"
},
{
"path": "dit.bin",
"size": 78688256,
"sha256": "d8e3c9c1347ee419712f10a65e49bdc87113b30d3cc0dec5d61c2d61ba9cebd6"
},
{
"path": "flow_encoder.onnx",
"size": 185649479,
"sha256": "6cf1a240eb98dfd1122c79e4c1ebd368fca84ebecad74a07029d12251f33efe9"
},
{
"path": "hift_decode.bin",
"size": 21569536,
"sha256": "c6837f86960ce01a1f151a8bf4723fd3a95c91c43ce7e7a7f237c8a12c2bb022"
},
{
"path": "hift_f0.bin",
"size": 3399680,
"sha256": "1fd234aae7cee68c1c4e3db8f9460f61b2d695f04dcb2a8478ba37a30a26f33a"
},
{
"path": "llm_head.bin",
"size": 626921472,
"sha256": "77588d705df56cb525f329149a1a97ac5c96a635b6338fb93c492f1461939328"
},
{
"path": "llm_part_0.bin",
"size": 926277632,
"sha256": "f4c34f35b6c7ab50dab27132486d90033f267915f4b2675c26d4593f69f026a0"
},
{
"path": "llm_part_1.bin",
"size": 926265344,
"sha256": "cb0642be52cfe18736d66807dc01c210acfba202988caf2b3fea1389c35652a6"
},
{
"path": "llm_part_2.bin",
"size": 926277632,
"sha256": "851d83fdd0c206176677ade08189b01bdfb4436525a04ebcbf010ab228156d28"
},
{
"path": "llm_part_3.bin",
"size": 926281728,
"sha256": "3eeb48f998a50f6de769bb4933977383aadf2f491d76bc3cb554aeb6e556c927"
},
{
"path": "tts_cpu.bin",
"size": 194742088,
"sha256": "74969808183cc9f91b37d0eee28c617a817b302af3ff445b98cbdea0e4301676"
},
{
"path": "vision_assets.bin",
"size": 11355264,
"sha256": "26af5e03e6ee94a8d595ad162c7a0c3fa8e169326f7224cb7868dd4fac240b3d"
},
{
"path": "vision.bin",
"size": 540069888,
"sha256": "a162ff53f8a49a2bd29839ae4fa0b44ed1814a1e28e7b1028dda2df9b09f1978"
}
]
}
]
},
{
"id": "moshiko",
"engine": "moshi",
"name": "Moshi · Moshiko",
"publisher": "Kyutai",
"summary": "7B 全双工语音对话模型(英文,实验性),Mimi 24 kHz 编解码,80 ms 一帧,边听边说。",
"languages": [
"en"
],
"license": "CC-BY-4.0",
"upstream": "kyutai/moshiko-pytorch-bf16",
"variants": [
{
"id": "moshiko-lm-v4-lpbq4-sm8850",
"title": "Temporal LPBQ W4 · DepFormer W8 · Mimi W8/W4",
"path": "moshi/moshiko-lm-v4-lpbq4-sm8850",
"socs": [
"SM8850"
],
"htp": "v81",
"qairt": "2.45.0.260326",
"notes": "QNN HTP context 绑定 SoC 与 QAIRT 版本;实测每帧凉机约 100 ms(RTF 1.25)、发热后约 130 ms(RTF 1.65),慢于实时,对话中延迟会逐渐累积;能回答开头一两个问题,约 45 秒后开始绕圈、不再回应。",
"files": [
{
"path": "frontend_embeddings.bin",
"size": 530718808,
"sha256": "2a2dc300998820869cd0a89e86d25f5e48753937d1f41df55b9dc1cb17d6100d"
},
{
"path": "mimi_encoder.bin",
"size": 128479232,
"sha256": "9fbff087226185d87ed3dc33cc8ccf8b644a34d17154a1188858d4e3c929b6c5"
},
{
"path": "mimi_decoder.bin",
"size": 53813248,
"sha256": "d20e273ecdb15e6d59d9e141ec96e81db4157e79116aef49ba6836ad366600c0"
},
{
"path": "dlc/head.bin",
"size": 131706880,
"sha256": "69d06212626ce728693bdc21dcf62dd70a4b3e2a77f63d10aa120cb28c0a1067"
},
{
"path": "dlc/depformer_codebook_0.bin",
"size": 150396928,
"sha256": "13b7d99126940df72757d7db16dea3578a9a143592ac5af115201ebbf749ab6f"
},
{
"path": "dlc/depformer_codebook_1.bin",
"size": 89055232,
"sha256": "e753db4c32e3b1c69e9162927d6e23f135d29ed2f0aa305c303dfb2486f2fee0"
},
{
"path": "dlc/depformer_codebook_2.bin",
"size": 89038848,
"sha256": "53fceb1e0c20a66d36854a919a77064890b79b84924ebb388ca01d9313189aeb"
},
{
"path": "dlc/depformer_codebook_3.bin",
"size": 89022464,
"sha256": "4f2a2f29d1433f0ecf8d4894d35a434158bb17417cc5f84b41a2cf6e99a1fa5e"
},
{
"path": "dlc/depformer_codebook_4.bin",
"size": 89026560,
"sha256": "75e658815ce90fb28ee73648f4917a874589370daddd30ad81f4a42efa84fdfa"
},
{
"path": "dlc/depformer_codebook_5.bin",
"size": 89022464,
"sha256": "290190f47716fa420dc607e3226fbec9fb4e61e9965a4fb831e8820693e32d37"
},
{
"path": "dlc/depformer_codebook_6.bin",
"size": 89030656,
"sha256": "cfd12a20bdfc338c2c0a19d284798803911dcc9cb486bfd2df7dfe3398bdccf4"
},
{
"path": "dlc/depformer_codebook_7.bin",
"size": 89022464,
"sha256": "7f76966b10d15ddc6449653462744fd7512357a0f19bcddbf7cefd8fe395affb"
},
{
"path": "dlc/temporal_part_0.bin",
"size": 859291648,
"sha256": "fc83d22b6e1caa027a3c12a3d456197ec9c09e9c32f4dc43310dd5f9da90bea3"
},
{
"path": "dlc/temporal_part_1.bin",
"size": 859312128,
"sha256": "97c8216ff9be49defee60f60b7b77bd87c84bdc7adf234eb513e2c067d982bca"
},
{
"path": "dlc/temporal_part_2.bin",
"size": 859287552,
"sha256": "608ef5b8fc26a1a1a6a2996224dc86e5a1eee4200b3aca84531139fe679f4cc2"
},
{
"path": "dlc/temporal_part_3.bin",
"size": 859308032,
"sha256": "e1b7e60f2d7e5f55981e241fb043ed83f57e39829127925dd8037fd346dc29dd"
},
{
"path": "text_vocab.txt",
"size": 296134,
"sha256": "3ffec0c8b77830b019735efe8844dff234a4fb24ae241cad3a8ca1475467eb6d"
}
]
}
]
}
]
}