{"text": "Unthorbit最强模型Cloud Fable 5全球金融后面一直都有人想去效仿它前有OpenRouter上线的Fusion模型这个我之前也分享过后有Transformer发明者Ion Jones两人共同创办的AI创业公司Sakuna AI才日本创建的所以起了日本名字所以说一款模型叫做Fugu Ultra性能据说是比肩Cloud Fable和Mythos的Fugu的声明中明确说无需承担出口管风险的牵沿能力也是针对Fable 5说的在行业最严格的工程科学推理基准测试中的Fugu跟Fable的对比模型看上去相差无几几乎是可以匹配的Mythos和Fable据说是一个单模型的架构Fugu思路有点使这些巧劲它的说法叫做集体智能革新逻辑就是背后编排了一整个可自由切换的AI的智能体池碰到单一供应商限制时能自动绕道什么叫限制呢就是被供应商没法实验它需要的能力换模型继续跑系统韧性是大幅提升的从那图中可以看出来Sakuna Fugu更像是一个编排的agent它编排对应的有一系列的模型包括了自己的模型也包括了开源和避源的模型它属于一个大模型的池子Fugu会动态的编排全球最顶尖的模型来完成复杂的多步骤任务所以它的核心的观点是未来比的不是谁的模型更大而谁能把全球的模型编排得更好更稳更自主表有意思是Fugu Ultra在SWE Bench Pro和Terminal Bench 2.1两个基准测试上都达到了当前的最优水平性能是明显的提升在科学推理方面Fugu模型依然也表现显著甚至超越了Methos Preview跟Fuble 5这印证了Fugu的一个核心的能力就是智能调度成为了提升性能的另一个维度并不依赖于增加更多的训练算理在补充的基准测试中Fugu和前沿的三个模型Germel 3.1 Pro HighOp4.8 MaxGPT 5.5X High做了利民的对比接受在这种情况他们的Fugu也是很难打我看了一下Fugu的运行机制非常意思它不是按照我们传统意义上的用人工提示词来编排工作流而是通过一个语言模型作为一个骨干基于Prompt上压文来生成Hidden State基于引擎状态再来协调其他的工作模型的池子就是说所有的编排工作其实是在Hidden State内实现的看了一下它的架构它的语言模型应该指的就是Sekona Fugu这个模型自己会生成一个Output同时在Logits之前的Hidden State又会路由出来传到它的工作池里面的所有的模型中再由这些模型去输出它的Logits如果能做到这一步的话应该它的所谓的工作池的模型都是开源模型因为必源模型是没法接触它的Hidden State作为Input来输出Logits所以它整个过程的设影也是比较巧妙不是单纯的靠Agent编排就能实现的一个路由逻辑Fugu在训练阶段是采用两步走第一步是大规模的广度的监督微调涵盖了编程数学推理语言理解等多方面第二步就是针对不同的单个任务进一步的监督微调实现端到端的优化这里面它们透露的一个点就是用了不同的编码助手就是Hardless的数据包括CodexCloud CodeOpenCode等收集了真实世界的多轮轨迹构建了设计仓务上要文迭代编程工具调用执行反馈和最终任务的多个端到端的任务用力虚拟", "segments": [{"id": 0, "seek": 0, "start": 0.0, "end": 2.82, "text": "Unthorbit最强模型Cloud Fable 5全球金融", "tokens": [50365, 12405, 392, 284, 5260, 8661, 5702, 118, 41908, 39823, 32787, 479, 712, 1025, 11319, 28533, 19117, 43772, 235, 50506], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 1, "seek": 0, "start": 2.82, "end": 4.86, "text": "后面一直都有人想去效仿它", "tokens": [50506, 13547, 8833, 34448, 7182, 45820, 7093, 6734, 43076, 1550, 123, 11284, 50608], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 2, "seek": 0, "start": 4.86, "end": 7.24, "text": "前有OpenRouter上线的Fusion模型", "tokens": [50608, 8945, 2412, 45569, 49, 23985, 5708, 16853, 123, 1546, 37, 5704, 41908, 39823, 50727], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 3, "seek": 0, "start": 7.24, "end": 8.24, "text": "这个我之前也分享过", "tokens": [50727, 15368, 1654, 32442, 6404, 30855, 16866, 50777], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 4, "seek": 0, "start": 8.24, "end": 10.4, "text": "后有Transformer发明者Ion Jones", "tokens": [50777, 13547, 2412, 33339, 837, 260, 28926, 11100, 12444, 40, 266, 10512, 50885], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 5, "seek": 0, "start": 10.4, "end": 13.34, "text": "两人共同创办的AI创业公司Sakuna AI", "tokens": [50885, 36257, 4035, 15408, 13089, 2437, 249, 39453, 1546, 48698, 2437, 249, 940, 248, 13545, 32981, 50, 514, 5051, 7318, 51032], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 6, "seek": 0, "start": 13.34, "end": 14.18, "text": "才日本创建的", "tokens": [51032, 18888, 27311, 2437, 249, 34157, 1546, 51074], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 7, "seek": 0, "start": 14.18, "end": 15.0, "text": "所以起了日本名字", "tokens": [51074, 7239, 9147, 2289, 27311, 15940, 22381, 51115], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 8, "seek": 0, "start": 15.0, "end": 17.2, "text": "所以说一款模型叫做Fugu Ultra", "tokens": [51115, 7239, 8090, 2257, 48798, 41908, 39823, 19855, 10907, 37, 13705, 20925, 51225], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 9, "seek": 0, "start": 17.2, "end": 20.16, "text": "性能据说是比肩Cloud Fable和Mythos的", "tokens": [51225, 21686, 8225, 26075, 106, 8090, 1541, 11706, 14356, 102, 32787, 479, 712, 12565, 8506, 392, 329, 1546, 51373], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 10, "seek": 0, "start": 20.16, "end": 22.02, "text": "Fugu的声明中明确说", "tokens": [51373, 37, 13705, 1546, 32045, 11100, 5975, 11100, 38114, 106, 8090, 51466], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 11, "seek": 0, "start": 22.02, "end": 24.6, "text": "无需承担出口管风险的牵沿能力", "tokens": [51466, 31364, 32535, 3416, 123, 6852, 227, 7781, 18144, 23131, 47209, 8842, 102, 1546, 6935, 113, 3308, 123, 8225, 13486, 51595], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 12, "seek": 0, "start": 24.6, "end": 26.080000000000002, "text": "也是针对Fable 5说的", "tokens": [51595, 22021, 25153, 230, 8713, 37, 712, 1025, 8090, 1546, 51669], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 13, "seek": 0, "start": 26.080000000000002, "end": 28.84, "text": "在行业最严格的工程科学推理基准测试中的", "tokens": [51669, 3581, 8082, 940, 248, 8661, 940, 98, 30921, 1546, 23323, 29649, 42091, 29618, 33597, 13876, 26008, 6336, 228, 11038, 233, 5233, 243, 5975, 1546, 51807], "temperature": 0, "avg_logprob": -0.19407833644321987, "compression_ratio": 1.1947115384615385, "no_speech_prob": 2.749547985125833e-11}, {"id": 14, "seek": 2884, "start": 28.84, "end": 32.2, "text": "Fugu跟Fable的对比模型看上去相差", "tokens": [50365, 37, 13705, 9678, 37, 712, 1546, 8713, 11706, 41908, 39823, 4200, 5708, 6734, 15106, 21679, 50533], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 15, "seek": 2884, "start": 32.2, "end": 34.08, "text": "无几几乎是可以匹配的", "tokens": [50533, 31364, 6336, 254, 6336, 254, 2930, 236, 1541, 6723, 9937, 117, 38846, 1546, 50627], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 16, "seek": 2884, "start": 34.08, "end": 37.26, "text": "Mythos和Fable据说是一个单模型的架构", "tokens": [50627, 8506, 392, 329, 12565, 37, 712, 26075, 106, 8090, 1541, 20182, 47446, 41908, 39823, 1546, 7360, 114, 7360, 226, 50786], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 17, "seek": 2884, "start": 37.26, "end": 39.32, "text": "Fugu思路有点使这些巧劲", "tokens": [50786, 37, 13705, 8870, 24658, 2412, 12579, 22982, 5562, 13824, 43605, 5087, 110, 50889], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 18, "seek": 2884, "start": 39.32, "end": 40.96, "text": "它的说法叫做集体智能", "tokens": [50889, 45224, 8090, 11148, 19855, 10907, 26020, 29485, 5094, 118, 8225, 50971], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 19, "seek": 2884, "start": 40.96, "end": 42.66, "text": "革新逻辑就是背后编排了", "tokens": [50971, 5363, 102, 12560, 2215, 119, 9830, 239, 5620, 46329, 13547, 38109, 244, 44647, 2289, 51056], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 20, "seek": 2884, "start": 42.66, "end": 45.34, "text": "一整个可自由切换的AI的智能体池", "tokens": [51056, 2257, 27662, 7549, 4429, 45233, 23632, 26075, 95, 1546, 48698, 1546, 5094, 118, 8225, 29485, 12800, 254, 51190], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 21, "seek": 2884, "start": 45.34, "end": 47.06, "text": "碰到单一供应商限制时", "tokens": [51190, 16337, 108, 4511, 47446, 2257, 3254, 249, 44297, 45581, 43446, 25491, 15729, 51276], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 22, "seek": 2884, "start": 47.06, "end": 47.980000000000004, "text": "能自动绕道", "tokens": [51276, 8225, 9722, 34961, 10115, 243, 6025, 51322], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 23, "seek": 2884, "start": 47.980000000000004, "end": 48.66, "text": "什么叫限制呢", "tokens": [51322, 10440, 19855, 43446, 25491, 6240, 51356], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 24, "seek": 2884, "start": 48.66, "end": 50.56, "text": "就是被供应商没法实验它需要的能力", "tokens": [51356, 5620, 23238, 3254, 249, 44297, 45581, 10062, 11148, 24726, 49657, 234, 11284, 35748, 1546, 8225, 13486, 51451], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 25, "seek": 2884, "start": 50.56, "end": 51.5, "text": "换模型继续跑", "tokens": [51451, 26075, 95, 41908, 39823, 10115, 100, 10115, 255, 32585, 51498], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 26, "seek": 2884, "start": 51.5, "end": 53.019999999999996, "text": "系统韧性是大幅提升的", "tokens": [51498, 25368, 10115, 253, 13665, 100, 21686, 1541, 3582, 3509, 227, 20949, 41670, 1546, 51574], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 27, "seek": 2884, "start": 53.019999999999996, "end": 54.0, "text": "从那图中可以看出来", "tokens": [51574, 35630, 4184, 3919, 122, 5975, 6723, 4200, 44561, 51623], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 28, "seek": 2884, "start": 54.0, "end": 56.8, "text": "Sakuna Fugu更像是一个编排的agent", "tokens": [51623, 50, 514, 5051, 479, 13705, 19002, 12760, 1541, 20182, 38109, 244, 44647, 1546, 559, 317, 51763], "temperature": 0, "avg_logprob": -0.11914052786650481, "compression_ratio": 1.2710526315789474, "no_speech_prob": 1.2316688467739478e-11}, {"id": 29, "seek": 5680, "start": 56.8, "end": 58.959999999999994, "text": "它编排对应的有一系列的模型", "tokens": [50365, 11284, 38109, 244, 44647, 8713, 44297, 1546, 32241, 25368, 43338, 1546, 41908, 39823, 50473], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 30, "seek": 5680, "start": 58.959999999999994, "end": 60.279999999999994, "text": "包括了自己的模型", "tokens": [50473, 41828, 2289, 17645, 1546, 41908, 39823, 50539], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 31, "seek": 5680, "start": 60.279999999999994, "end": 61.919999999999995, "text": "也包括了开源和避源的模型", "tokens": [50539, 6404, 41828, 2289, 18937, 47402, 12565, 3330, 123, 47402, 1546, 41908, 39823, 50621], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 32, "seek": 5680, "start": 61.919999999999995, "end": 63.76, "text": "它属于一个大模型的池子", "tokens": [50621, 11284, 9636, 252, 37732, 20182, 3582, 41908, 39823, 1546, 12800, 254, 7626, 50713], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 33, "seek": 5680, "start": 63.76, "end": 66.16, "text": "Fugu会动态的编排全球最顶尖的模型", "tokens": [50713, 37, 13705, 12949, 34961, 3757, 223, 1546, 38109, 244, 44647, 11319, 28533, 8661, 10178, 114, 1530, 244, 1546, 41908, 39823, 50833], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 34, "seek": 5680, "start": 66.16, "end": 67.96, "text": "来完成复杂的多步骤任务", "tokens": [50833, 6912, 41509, 1787, 235, 4422, 224, 1546, 6392, 31429, 49657, 97, 26443, 5087, 94, 50923], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 35, "seek": 5680, "start": 67.96, "end": 69.75999999999999, "text": "所以它的核心的观点是", "tokens": [50923, 7239, 45224, 8987, 116, 7945, 1546, 42350, 12579, 1541, 51013], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 36, "seek": 5680, "start": 69.75999999999999, "end": 71.84, "text": "未来比的不是谁的模型更大", "tokens": [51013, 29954, 6912, 11706, 1546, 7296, 33556, 1546, 41908, 39823, 19002, 3582, 51117], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 37, "seek": 5680, "start": 71.84, "end": 73.75999999999999, "text": "而谁能把全球的模型编排得更好", "tokens": [51117, 11070, 33556, 8225, 16075, 11319, 28533, 1546, 41908, 39823, 38109, 244, 44647, 5916, 19002, 2131, 51213], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 38, "seek": 5680, "start": 73.75999999999999, "end": 74.44, "text": "更稳更自主", "tokens": [51213, 19002, 10415, 111, 19002, 9722, 13557, 51247], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 39, "seek": 5680, "start": 74.44, "end": 76.03999999999999, "text": "表有意思是Fugu Ultra", "tokens": [51247, 17571, 2412, 16697, 1541, 37, 13705, 20925, 51327], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 40, "seek": 5680, "start": 76.03999999999999, "end": 78.8, "text": "在SWE Bench Pro和Terminal Bench 2.1", "tokens": [51327, 3581, 50, 37937, 3964, 339, 1705, 12565, 51, 966, 2071, 3964, 339, 568, 13, 16, 51465], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 41, "seek": 5680, "start": 78.8, "end": 79.72, "text": "两个基准测试上", "tokens": [51465, 36257, 7549, 26008, 6336, 228, 11038, 233, 5233, 243, 5708, 51511], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 42, "seek": 5680, "start": 79.72, "end": 81.44, "text": "都达到了当前的最优水平", "tokens": [51511, 7182, 9830, 122, 21381, 16233, 8945, 1546, 8661, 7384, 246, 15590, 16716, 51597], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 43, "seek": 5680, "start": 81.44, "end": 82.75999999999999, "text": "性能是明显的提升", "tokens": [51597, 21686, 8225, 1541, 11100, 1431, 122, 1546, 20949, 41670, 51663], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 44, "seek": 5680, "start": 82.75999999999999, "end": 83.88, "text": "在科学推理方面", "tokens": [51663, 3581, 42091, 29618, 33597, 13876, 39604, 51719], "temperature": 0, "avg_logprob": -0.1686294338714455, "compression_ratio": 1.3342318059299192, "no_speech_prob": 1.5218963242413786e-11}, {"id": 45, "seek": 8388, "start": 83.88, "end": 86.11999999999999, "text": "Fugu模型依然也表现显著", "tokens": [50365, 37, 13705, 41908, 39823, 3254, 251, 5823, 6404, 17571, 20204, 1431, 122, 19382, 50477], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 46, "seek": 8388, "start": 86.11999999999999, "end": 88.32, "text": "甚至超越了Methos Preview跟Fuble 5", "tokens": [50477, 17556, 20844, 19869, 25761, 2289, 44, 3293, 329, 6001, 1759, 9678, 37, 84, 638, 1025, 50587], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 47, "seek": 8388, "start": 88.32, "end": 90.56, "text": "这印证了Fugu的一个核心的能力", "tokens": [50587, 5562, 35825, 5233, 223, 2289, 37, 13705, 1546, 20182, 8987, 116, 7945, 1546, 8225, 13486, 50699], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 48, "seek": 8388, "start": 90.56, "end": 91.39999999999999, "text": "就是智能调度", "tokens": [50699, 5620, 5094, 118, 8225, 8897, 225, 13127, 50741], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 49, "seek": 8388, "start": 91.39999999999999, "end": 93.11999999999999, "text": "成为了提升性能的另一个维度", "tokens": [50741, 11336, 13992, 2289, 20949, 41670, 21686, 8225, 1546, 22762, 20182, 10115, 112, 13127, 50827], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 50, "seek": 8388, "start": 93.11999999999999, "end": 94.88, "text": "并不依赖于增加更多的训练算理", "tokens": [50827, 3509, 114, 1960, 3254, 251, 5266, 244, 37732, 24228, 252, 9990, 19002, 6392, 1546, 7422, 255, 10115, 225, 19497, 13876, 50915], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 51, "seek": 8388, "start": 94.88, "end": 96.16, "text": "在补充的基准测试中", "tokens": [50915, 3581, 9890, 98, 2347, 227, 1546, 26008, 6336, 228, 11038, 233, 5233, 243, 5975, 50979], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 52, "seek": 8388, "start": 96.16, "end": 98.44, "text": "Fugu和前沿的三个模型", "tokens": [50979, 37, 13705, 12565, 8945, 3308, 123, 1546, 10960, 7549, 41908, 39823, 51093], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 53, "seek": 8388, "start": 98.44, "end": 99.8, "text": "Germel 3.1 Pro High", "tokens": [51093, 38, 966, 338, 805, 13, 16, 1705, 5229, 51161], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 54, "seek": 8388, "start": 99.8, "end": 100.75999999999999, "text": "Op4.8 Max", "tokens": [51161, 46, 79, 19, 13, 23, 7402, 51209], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 55, "seek": 8388, "start": 100.75999999999999, "end": 101.64, "text": "GPT 5.5", "tokens": [51209, 38, 47, 51, 1025, 13, 20, 51253], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 56, "seek": 8388, "start": 101.64, "end": 102.03999999999999, "text": "X High", "tokens": [51253, 55, 5229, 51273], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 57, "seek": 8388, "start": 102.03999999999999, "end": 103.0, "text": "做了利民的对比", "tokens": [51273, 10907, 2289, 23700, 16113, 1546, 8713, 11706, 51321], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 58, "seek": 8388, "start": 103.0, "end": 103.64, "text": "接受在这种情况", "tokens": [51321, 14468, 23151, 3581, 5562, 39810, 46514, 51353], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 59, "seek": 8388, "start": 103.64, "end": 105.36, "text": "他们的Fugu也是很难打", "tokens": [51353, 47911, 1546, 37, 13705, 22021, 4563, 46531, 12467, 51439], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 60, "seek": 8388, "start": 105.36, "end": 107.44, "text": "我看了一下Fugu的运行机制", "tokens": [51439, 1654, 4200, 2289, 8861, 37, 13705, 1546, 3316, 238, 8082, 37960, 25491, 51543], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 61, "seek": 8388, "start": 107.44, "end": 108.03999999999999, "text": "非常意思", "tokens": [51543, 14392, 16697, 51573], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 62, "seek": 8388, "start": 108.03999999999999, "end": 110.67999999999999, "text": "它不是按照我们传统意义上的", "tokens": [51573, 11284, 7296, 26613, 32150, 15003, 7384, 254, 10115, 253, 9042, 2930, 231, 5708, 1546, 51705], "temperature": 0, "avg_logprob": -0.1644889592055248, "compression_ratio": 1.183168316831683, "no_speech_prob": 1.4787027505236416e-11}, {"id": 63, "seek": 11068, "start": 110.68, "end": 113.16000000000001, "text": "用人工提示词来编排工作流", "tokens": [50365, 9254, 4035, 23323, 20949, 25696, 5233, 235, 6912, 38109, 244, 44647, 41315, 27854, 50489], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 64, "seek": 11068, "start": 113.16000000000001, "end": 115.24000000000001, "text": "而是通过一个语言模型", "tokens": [50489, 11070, 1541, 19550, 16866, 20182, 5233, 255, 12009, 41908, 39823, 50593], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 65, "seek": 11068, "start": 115.24000000000001, "end": 116.32000000000001, "text": "作为一个骨干", "tokens": [50593, 11914, 13992, 20182, 165, 16838, 26111, 50647], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 66, "seek": 11068, "start": 116.32000000000001, "end": 117.56, "text": "基于Prompt上压文", "tokens": [50647, 26008, 37732, 47, 4397, 662, 5708, 5014, 233, 17174, 50709], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 67, "seek": 11068, "start": 117.56, "end": 119.16000000000001, "text": "来生成Hidden State", "tokens": [50709, 6912, 8244, 11336, 39, 6171, 4533, 50789], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 68, "seek": 11068, "start": 119.16000000000001, "end": 119.92, "text": "基于引擎状态", "tokens": [50789, 26008, 37732, 41301, 13167, 236, 35276, 114, 3757, 223, 50827], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 69, "seek": 11068, "start": 119.92, "end": 122.4, "text": "再来协调其他的工作模型的池子", "tokens": [50827, 8623, 6912, 5322, 237, 8897, 225, 9572, 31309, 41315, 41908, 39823, 1546, 12800, 254, 7626, 50951], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 70, "seek": 11068, "start": 122.4, "end": 123.96000000000001, "text": "就是说所有的编排工作", "tokens": [50951, 5620, 8090, 39300, 1546, 38109, 244, 44647, 41315, 51029], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 71, "seek": 11068, "start": 123.96000000000001, "end": 126.32000000000001, "text": "其实是在Hidden State内实现的", "tokens": [51029, 41646, 1541, 3581, 39, 6171, 4533, 34742, 24726, 20204, 1546, 51147], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 72, "seek": 11068, "start": 126.32000000000001, "end": 127.08000000000001, "text": "看了一下它的架构", "tokens": [51147, 4200, 2289, 8861, 45224, 7360, 114, 7360, 226, 51185], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 73, "seek": 11068, "start": 127.08000000000001, "end": 128.04000000000002, "text": "它的语言模型", "tokens": [51185, 45224, 5233, 255, 12009, 41908, 39823, 51233], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 74, "seek": 11068, "start": 128.04000000000002, "end": 129.84, "text": "应该指的就是Sekona Fugu", "tokens": [51233, 44297, 44646, 25922, 1546, 5620, 50, 916, 4037, 479, 13705, 51323], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 75, "seek": 11068, "start": 129.84, "end": 130.36, "text": "这个模型", "tokens": [51323, 15368, 41908, 39823, 51349], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 76, "seek": 11068, "start": 130.36, "end": 131.48000000000002, "text": "自己会生成一个Output", "tokens": [51349, 17645, 12949, 8244, 11336, 20182, 28353, 2582, 51405], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 77, "seek": 11068, "start": 131.48000000000002, "end": 133.84, "text": "同时在Logits之前的Hidden State", "tokens": [51405, 13089, 15729, 3581, 43, 664, 1208, 32442, 1546, 39, 6171, 4533, 51523], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 78, "seek": 11068, "start": 133.84, "end": 135.0, "text": "又会路由出来", "tokens": [51523, 17047, 12949, 24658, 23786, 44561, 51581], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 79, "seek": 11068, "start": 135.0, "end": 136.60000000000002, "text": "传到它的工作池里面的", "tokens": [51581, 7384, 254, 4511, 45224, 41315, 12800, 254, 15759, 8833, 1546, 51661], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 80, "seek": 11068, "start": 136.60000000000002, "end": 137.60000000000002, "text": "所有的模型中", "tokens": [51661, 39300, 1546, 41908, 39823, 5975, 51711], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 81, "seek": 11068, "start": 137.60000000000002, "end": 138.44, "text": "再由这些模型", "tokens": [51711, 8623, 23786, 5562, 13824, 41908, 39823, 51753], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 82, "seek": 11068, "start": 138.44, "end": 140.12, "text": "去输出它的Logits", "tokens": [51753, 6734, 9830, 241, 7781, 45224, 43, 664, 1208, 51837], "temperature": 0, "avg_logprob": -0.12421790413234544, "compression_ratio": 1.424929178470255, "no_speech_prob": 1.5495674188237274e-11}, {"id": 83, "seek": 14012, "start": 140.12, "end": 141.64000000000001, "text": "如果能做到这一步的话", "tokens": [50365, 13119, 8225, 10907, 4511, 5562, 2257, 31429, 44575, 50441], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 84, "seek": 14012, "start": 141.64000000000001, "end": 143.84, "text": "应该它的所谓的工作池的模型", "tokens": [50441, 44297, 44646, 45224, 5966, 8897, 241, 1546, 41315, 12800, 254, 1546, 41908, 39823, 50551], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 85, "seek": 14012, "start": 143.84, "end": 144.68, "text": "都是开源模型", "tokens": [50551, 22796, 18937, 47402, 41908, 39823, 50593], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 86, "seek": 14012, "start": 144.68, "end": 145.44, "text": "因为必源模型", "tokens": [50593, 34627, 28531, 47402, 41908, 39823, 50631], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 87, "seek": 14012, "start": 145.44, "end": 147.32, "text": "是没法接触它的Hidden State", "tokens": [50631, 1541, 10062, 11148, 14468, 6758, 99, 45224, 39, 6171, 4533, 50725], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 88, "seek": 14012, "start": 147.32, "end": 149.12, "text": "作为Input来输出Logits", "tokens": [50725, 11914, 13992, 4575, 2582, 6912, 9830, 241, 7781, 43, 664, 1208, 50815], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 89, "seek": 14012, "start": 149.12, "end": 150.32, "text": "所以它整个过程的设影", "tokens": [50815, 7239, 11284, 27662, 7549, 16866, 29649, 1546, 7422, 122, 16820, 50875], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 90, "seek": 14012, "start": 150.32, "end": 151.04, "text": "也是比较巧妙", "tokens": [50875, 22021, 11706, 9830, 225, 43605, 5648, 247, 50911], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 91, "seek": 14012, "start": 151.04, "end": 153.36, "text": "不是单纯的靠Agent编排", "tokens": [50911, 7296, 47446, 16853, 107, 1546, 5363, 254, 32, 6930, 38109, 244, 44647, 51027], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 92, "seek": 14012, "start": 153.36, "end": 155.48000000000002, "text": "就能实现的一个路由逻辑", "tokens": [51027, 3111, 8225, 24726, 20204, 1546, 20182, 24658, 23786, 2215, 119, 9830, 239, 51133], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 93, "seek": 14012, "start": 155.48000000000002, "end": 156.6, "text": "Fugu在训练阶段", "tokens": [51133, 37, 13705, 3581, 7422, 255, 10115, 225, 10034, 114, 28427, 51189], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 94, "seek": 14012, "start": 156.6, "end": 157.6, "text": "是采用两步走", "tokens": [51189, 1541, 5873, 229, 9254, 36257, 31429, 9575, 51239], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 95, "seek": 14012, "start": 157.6, "end": 160.04000000000002, "text": "第一步是大规模的广度的监督微调", "tokens": [51239, 18049, 31429, 1541, 3582, 6758, 226, 41908, 1546, 3509, 123, 13127, 1546, 5419, 239, 14221, 96, 39152, 8897, 225, 51361], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 96, "seek": 14012, "start": 160.04000000000002, "end": 161.0, "text": "涵盖了编程", "tokens": [51361, 35681, 113, 5419, 244, 2289, 38109, 244, 29649, 51409], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 97, "seek": 14012, "start": 161.0, "end": 161.88, "text": "数学推理", "tokens": [51409, 33188, 29618, 33597, 13876, 51453], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 98, "seek": 14012, "start": 161.88, "end": 162.4, "text": "语言理解", "tokens": [51453, 5233, 255, 12009, 13876, 17278, 51479], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 99, "seek": 14012, "start": 162.4, "end": 162.96, "text": "等多方面", "tokens": [51479, 10187, 6392, 39604, 51507], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 100, "seek": 14012, "start": 162.96, "end": 165.36, "text": "第二步就是针对不同的单个任务", "tokens": [51507, 19693, 31429, 5620, 25153, 230, 8713, 47123, 1546, 47446, 7549, 26443, 5087, 94, 51627], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 101, "seek": 14012, "start": 165.36, "end": 166.20000000000002, "text": "进一步的监督微调", "tokens": [51627, 36700, 2257, 31429, 1546, 5419, 239, 14221, 96, 39152, 8897, 225, 51669], "temperature": 0, "avg_logprob": -0.11437588849944384, "compression_ratio": 1.2898172323759791, "no_speech_prob": 1.587385604906455e-11}, {"id": 102, "seek": 16620, "start": 166.2, "end": 167.44, "text": "实现端到端的优化", "tokens": [50365, 24726, 20204, 11957, 107, 4511, 11957, 107, 1546, 7384, 246, 23756, 50427], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 103, "seek": 16620, "start": 167.44, "end": 168.04, "text": "这里面", "tokens": [50427, 35102, 8833, 50457], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 104, "seek": 16620, "start": 168.04, "end": 169.04, "text": "它们透露的一个点", "tokens": [50457, 11284, 9497, 45808, 18594, 110, 1546, 20182, 12579, 50507], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 105, "seek": 16620, "start": 169.04, "end": 170.76, "text": "就是用了不同的编码助手", "tokens": [50507, 5620, 9254, 2289, 47123, 1546, 38109, 244, 23230, 223, 37618, 11389, 50593], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 106, "seek": 16620, "start": 170.76, "end": 171.76, "text": "就是Hardless的数据", "tokens": [50593, 5620, 39, 515, 1832, 1546, 33188, 26075, 106, 50643], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 107, "seek": 16620, "start": 171.76, "end": 172.51999999999998, "text": "包括Codex", "tokens": [50643, 41828, 34, 1429, 87, 50681], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 108, "seek": 16620, "start": 172.51999999999998, "end": 173.0, "text": "Cloud Code", "tokens": [50681, 32787, 15549, 50705], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 109, "seek": 16620, "start": 173.0, "end": 173.76, "text": "OpenCode等", "tokens": [50705, 45569, 34, 1429, 10187, 50743], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 110, "seek": 16620, "start": 173.76, "end": 176.0, "text": "收集了真实世界的多轮轨迹", "tokens": [50743, 18681, 26020, 2289, 6303, 24726, 24486, 1546, 6392, 17819, 106, 17819, 101, 3316, 117, 50855], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 111, "seek": 16620, "start": 176.0, "end": 177.88, "text": "构建了设计仓务上要文", "tokens": [50855, 7360, 226, 34157, 2289, 7422, 122, 7422, 94, 1550, 241, 5087, 94, 5708, 4275, 17174, 50949], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 112, "seek": 16620, "start": 177.88, "end": 178.56, "text": "迭代编程", "tokens": [50949, 3316, 255, 19105, 38109, 244, 29649, 50983], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 113, "seek": 16620, "start": 178.56, "end": 179.11999999999998, "text": "工具调用", "tokens": [50983, 23323, 39806, 8897, 225, 9254, 51011], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 114, "seek": 16620, "start": 179.11999999999998, "end": 179.67999999999998, "text": "执行反馈", "tokens": [51011, 3416, 100, 8082, 22138, 11748, 230, 51039], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 115, "seek": 16620, "start": 179.67999999999998, "end": 180.56, "text": "和最终任务的", "tokens": [51039, 12565, 8661, 10115, 230, 26443, 5087, 94, 1546, 51083], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}, {"id": 116, "seek": 16620, "start": 180.56, "end": 182.0, "text": "多个端到端的任务用力虚拟", "tokens": [51083, 6392, 7549, 11957, 107, 4511, 11957, 107, 1546, 26443, 5087, 94, 9254, 13486, 12026, 248, 6852, 253, 51155], "temperature": 0, "avg_logprob": -0.18265740494979055, "compression_ratio": 1.1488549618320612, "no_speech_prob": 1.5820003293476326e-11}], "language": "zh"}