{"text": "写出Redis的那个人刚刚又做了一个自己的AI引擎他居然能在你面前那台笔记本上跑起接近前沿模型的效果这个项目叫DS4不过引擎本身有个更好听的名字叫Dwarf Star作者是Salvatore Sanfilippo也就是大家熟知的Antares Redis的创始人大约两个月前他用C从0写了一个全新的推理引擎只为了让DeepSeek的V4模型能在你可能真的拥有的机器上跑得很好比如MacBook或MacStudio接下来的惊喜是模型不再非得塞进内存里它把你的SSD也当成它真正的家需要时再从硬盘流逝读取那些重的部分而不是逼你把整个模型都塞进人到现在为止还喜欢吗订阅这个YouTube频道还可以去Indie Hacker News免费领取每日Newsletter每个工作日都有一个新的深度解析如果你真的打开这个Repo最显眼的就是他对这一切都很坦诚开头他就说了这个项目故意做得很窄一次只针对一个模型而且他不是想做通用运行器往下再看一点你会看到他的动机基本上就是真正有能力的开源模型终于和真正有能力的个人电脑同时出现了所以这两者该碰到一起了继续往下你会看到真正的速度数据都是在真实的MacBook上跑出来的实际Benchmark还有那些小图表就算上下文越来越长Token Per Second也几乎没怎么掉然后他会讲到怎么运行比你内存还大的模型快到下面时最疯狂的部分来了把一个模型拆到两台用网线连在一起的笔记本上跑说实话在自己机器上跑模型也不新鲜了有Lama CPP这个庞大的开源项目整个生态里有一半都是建立在他之上Dwarf Star也毫不掩饰地站在他肩膀上Alama把本地模型变成了一型命令安装LM Studio又把这一切包装成了友好的桌面应用连不爱用终端的人也能上手但这个项目不一样他走的是完全相反的路他不是什么模型都支持而是只选一个然后把它调到极致拼命榨出最后一点速度和效果先说下这个规模已经有18500多个Star了而且是MIT许可证所以你基本可以随便拿去用整个项目几乎完全用C写成真想跑得顺还是得有一台差不多96G内存的Mac当然内存更少也能流逝运行好机制上它到底是怎么工作的核心是在现代机器上你的模型和缓存不一定非得放在RAM里也可以放在SSD上这个思路一变事情就完全不一样了路由专家层也就是这类Mixture Model里占大头的部分会在某个Token需要它的时候才从磁盘流逝加赛这样一来内存就不再是一睹你一头撞上的硬墙而更像是加速用的杠杆为了让它能跑起来它用了比较激进的2bit量化但这是有意做成不对称的它只压缩Exper权重Attention和共享层则尽量保持接近满质量所以它还能保持足够稳定值得拿来做真实任务而结果确实有效在M3MAX笔记本上速度大约是每秒27个Token已经完全够用了想被推荐吗把你的开源仓库提交到Indie HackerNew Supply就能让成千上万的开发者看到你说实话到这里它就不再只是个技术演示了而是开始像个真正的产品了它内置了一个编码Agent有个很巧的设计师这个绘画不会存到什么数据库里这个Session本质上就是把KVCache放在你的硬盘上这样切换以保存的对话就会非常快它还内置了一个同时支持OpenLayer和Enthropy API的Server这意味着你可以直接把Codex或Cloud Code指向你自己的笔记本电脑在你自己托管的模型上工作它甚至加了一个PowerDial可以在你更在意机器两款安静而不是榨干每一个Token的时候去限制GPU所有这些都在指向同一个想法本地模型应该像一个做完整的产品而不是一堆零件说到底它的底层其实非常老派核心部分是纯C再加上一些手写的Cuda和Metal Shader真正负责GPU运算的几乎就只有这些它明确没有链接GGML也就是LamaCPP背后的那个库但它确实适配了其中一些格式和Kernel而且出于尊重把它们的版权信息直接保留了下来每次构建都会拿DivC在不同上下文长度下的官方输出做检查所以那种悄悄把质量搞坏的改动会在发布前被发现它还能跑在Apple Metal和NVIDIA显卡上连小小的DGX2也支持它甚至还能处理AMD的统一内存机器启动它出人意料的简单甚至有点无聊但这是好事你先运行下载脚本把适配你机器的2bit模型拉下来然后在Mac上用Mac构建Metal版本再把它启动起来你还可以选择开启SSD Streaming这样就能用比实际内存更大的模型之后你就是在和自己硬件上的模型对话输入的内容也不会离开这台机器然后还有一个最近才上线让大家都有点惊讶的功能Distributed Mode它会把一个模型拆到多台机器上每台分一部分layers这样一个大到任何单机都跑不动的模型也照样能运行最典型的例子是两台128G的MacBook通过Thunderbolt连接在它们之间运行完整的4bit模型更惊人的是这一切都是在最近几周的提交里才出现的所以这对你意味着什么现在还很早作者也首先承认这还是Beta已经有几百个未解决的问题而且它只会运行它发布的那些特定模型文件所以它是把利器不是瑞士军刀你还得有一台相当强的Mac才能获得好的体验也有些人会介意这里面很多代码都是在AI参与下写出来的但往后看这背后的压注其实挺让人兴奋的它在说这么强大的模型完全跑在你自己的硬件上而且不会把任何东西回传出去这不是以后才会有的事而是现在就能做到的事如果你是那种想把整套技术站都摆在桌面上而不是放在别人云端里的开发者这大概就是目前最清楚的信号说明局面正在变化再想到这位作者下一次他发出来的东西肯定也会很有意思值得一看", "segments": [{"id": 0, "seek": 0, "start": 0.0, "end": 3.54, "text": "写出Redis的那个人刚刚又做了一个自己的AI引擎", "tokens": [50365, 5676, 247, 7781, 20544, 271, 1546, 43083, 4035, 49160, 49160, 17047, 10907, 2289, 20182, 17645, 1546, 48698, 41301, 13167, 236, 50542], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 1, "seek": 0, "start": 3.54, "end": 6.38, "text": "他居然能在你面前那台笔记本上", "tokens": [50542, 5000, 9636, 227, 5823, 8225, 3581, 2166, 8833, 8945, 4184, 15433, 5437, 242, 34756, 8802, 5708, 50684], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 2, "seek": 0, "start": 6.38, "end": 8.66, "text": "跑起接近前沿模型的效果", "tokens": [50684, 32585, 9147, 14468, 17463, 8945, 3308, 123, 41908, 39823, 1546, 43076, 9319, 50798], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 3, "seek": 0, "start": 8.66, "end": 10.52, "text": "这个项目叫DS4", "tokens": [50798, 15368, 10178, 117, 11386, 19855, 11844, 19, 50891], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 4, "seek": 0, "start": 10.52, "end": 13.0, "text": "不过引擎本身有个更好听的名字", "tokens": [50891, 1960, 16866, 41301, 13167, 236, 8802, 19847, 2412, 7549, 19002, 2131, 31022, 1546, 15940, 22381, 51015], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 5, "seek": 0, "start": 13.0, "end": 14.700000000000001, "text": "叫Dwarf Star", "tokens": [51015, 19855, 35, 6925, 69, 5705, 51100], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 6, "seek": 0, "start": 14.700000000000001, "end": 17.400000000000002, "text": "作者是Salvatore Sanfilippo", "tokens": [51100, 11914, 12444, 1541, 39013, 23352, 418, 5271, 69, 12833, 78, 51235], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 7, "seek": 0, "start": 17.400000000000002, "end": 21.46, "text": "也就是大家熟知的Antares Redis的创始人", "tokens": [51235, 6404, 5620, 6868, 21805, 253, 6498, 1546, 35807, 8643, 4477, 271, 1546, 2437, 249, 15476, 4035, 51438], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 8, "seek": 0, "start": 21.46, "end": 23.22, "text": "大约两个月前", "tokens": [51438, 3582, 16853, 99, 36257, 7549, 6939, 8945, 51526], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 9, "seek": 0, "start": 23.22, "end": 26.12, "text": "他用C从0写了一个全新的推理引擎", "tokens": [51526, 5000, 9254, 34, 35630, 15, 5676, 247, 2289, 20182, 11319, 12560, 1546, 33597, 13876, 41301, 13167, 236, 51671], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 10, "seek": 0, "start": 26.12, "end": 29.04, "text": "只为了让DeepSeek的V4模型", "tokens": [51671, 14003, 13992, 2289, 33650, 11089, 595, 10637, 916, 1546, 53, 19, 41908, 39823, 51817], "temperature": 0, "avg_logprob": -0.14609066056616513, "compression_ratio": 1.165644171779141, "no_speech_prob": 1.7780657154964352e-11}, {"id": 11, "seek": 2904, "start": 29.04, "end": 32.04, "text": "能在你可能真的拥有的机器上跑得很好", "tokens": [50365, 8225, 3581, 2166, 16657, 8034, 6852, 98, 2412, 1546, 37960, 34386, 5708, 32585, 5916, 23801, 50515], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 12, "seek": 2904, "start": 32.04, "end": 34.62, "text": "比如MacBook或MacStudio", "tokens": [50515, 36757, 34355, 19203, 19780, 34355, 42665, 1004, 50644], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 13, "seek": 2904, "start": 34.62, "end": 36.22, "text": "接下来的惊喜是", "tokens": [50644, 14468, 4438, 6912, 1546, 4799, 232, 13620, 1541, 50724], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 14, "seek": 2904, "start": 36.22, "end": 38.72, "text": "模型不再非得塞进内存里", "tokens": [50724, 41908, 39823, 1960, 8623, 12107, 5916, 13331, 252, 36700, 34742, 39781, 15759, 50849], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 15, "seek": 2904, "start": 38.72, "end": 41.879999999999995, "text": "它把你的SSD也当成它真正的家", "tokens": [50849, 11284, 16075, 18961, 21929, 35, 6404, 16233, 11336, 11284, 6303, 15789, 1546, 5155, 51007], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 16, "seek": 2904, "start": 41.879999999999995, "end": 45.64, "text": "需要时再从硬盘流逝读取那些重的部分", "tokens": [51007, 35748, 15729, 8623, 35630, 38114, 105, 5419, 246, 27854, 2215, 251, 5233, 119, 29436, 4184, 13824, 12624, 1546, 32174, 51195], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 17, "seek": 2904, "start": 45.64, "end": 48.68, "text": "而不是逼你把整个模型都塞进人", "tokens": [51195, 11070, 7296, 2215, 120, 2166, 16075, 27662, 7549, 41908, 39823, 7182, 13331, 252, 36700, 4035, 51347], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 18, "seek": 2904, "start": 48.68, "end": 50.739999999999995, "text": "到现在为止还喜欢吗", "tokens": [51347, 4511, 25040, 13992, 30438, 14852, 41606, 14769, 51450], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 19, "seek": 2904, "start": 50.739999999999995, "end": 52.4, "text": "订阅这个YouTube频道", "tokens": [51450, 7422, 95, 10034, 227, 15368, 3223, 41173, 39752, 6025, 51533], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 20, "seek": 2904, "start": 52.4, "end": 54.72, "text": "还可以去Indie Hacker News", "tokens": [51533, 14852, 6723, 6734, 21790, 414, 389, 23599, 7987, 51649], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 21, "seek": 2904, "start": 54.72, "end": 56.34, "text": "免费领取每日Newsletter", "tokens": [51649, 2347, 235, 18464, 117, 12501, 228, 29436, 23664, 6890, 35766, 21248, 51730], "temperature": 0, "avg_logprob": -0.12727450035713814, "compression_ratio": 1.1415384615384616, "no_speech_prob": 9.636872896900961e-12}, {"id": 22, "seek": 5634, "start": 56.34, "end": 59.38, "text": "每个工作日都有一个新的深度解析", "tokens": [50365, 23664, 7549, 41315, 6890, 48121, 20182, 12560, 1546, 24043, 13127, 17278, 7360, 238, 50517], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 23, "seek": 5634, "start": 59.38, "end": 61.68000000000001, "text": "如果你真的打开这个Repo", "tokens": [50517, 45669, 8034, 12467, 18937, 15368, 25554, 78, 50632], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 24, "seek": 5634, "start": 61.68000000000001, "end": 65.0, "text": "最显眼的就是他对这一切都很坦诚", "tokens": [50632, 8661, 1431, 122, 25281, 1546, 5620, 5000, 8713, 5562, 2257, 23632, 7182, 4563, 14872, 99, 5233, 248, 50798], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 25, "seek": 5634, "start": 65.0, "end": 66.5, "text": "开头他就说了", "tokens": [50798, 18937, 39862, 5000, 3111, 8090, 2289, 50873], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 26, "seek": 5634, "start": 66.5, "end": 68.74000000000001, "text": "这个项目故意做得很窄", "tokens": [50873, 15368, 10178, 117, 11386, 43045, 9042, 10907, 5916, 4563, 30771, 226, 50985], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 27, "seek": 5634, "start": 68.74000000000001, "end": 70.46000000000001, "text": "一次只针对一个模型", "tokens": [50985, 27505, 14003, 25153, 230, 8713, 20182, 41908, 39823, 51071], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 28, "seek": 5634, "start": 70.46000000000001, "end": 73.12, "text": "而且他不是想做通用运行器", "tokens": [51071, 22942, 5000, 7296, 7093, 10907, 19550, 9254, 3316, 238, 8082, 34386, 51204], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 29, "seek": 5634, "start": 73.12, "end": 74.30000000000001, "text": "往下再看一点", "tokens": [51204, 29510, 4438, 8623, 4200, 49042, 51263], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 30, "seek": 5634, "start": 74.30000000000001, "end": 76.10000000000001, "text": "你会看到他的动机", "tokens": [51263, 2166, 12949, 18032, 31309, 34961, 37960, 51353], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 31, "seek": 5634, "start": 76.10000000000001, "end": 79.06, "text": "基本上就是真正有能力的开源模型", "tokens": [51353, 37946, 5708, 5620, 6303, 15789, 2412, 8225, 13486, 1546, 18937, 47402, 41908, 39823, 51501], "temperature": 0, "avg_logprob": -0.12773534630527014, "compression_ratio": 1.2509803921568627, "no_speech_prob": 1.8930381567860977e-11}, {"id": 32, "seek": 7906, "start": 79.06, "end": 82.82000000000001, "text": "终于和真正有能力的个人电脑同时出现了", "tokens": [50365, 10115, 230, 37732, 12565, 6303, 15789, 2412, 8225, 13486, 1546, 7549, 4035, 42182, 27067, 239, 13089, 15729, 7781, 20204, 2289, 50553], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 33, "seek": 7906, "start": 82.82000000000001, "end": 85.48, "text": "所以这两者该碰到一起了", "tokens": [50553, 7239, 5562, 36257, 12444, 44646, 16337, 108, 4511, 29567, 2289, 50686], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 34, "seek": 7906, "start": 85.48, "end": 86.76, "text": "继续往下", "tokens": [50686, 10115, 100, 10115, 255, 29510, 4438, 50750], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 35, "seek": 7906, "start": 86.76, "end": 88.78, "text": "你会看到真正的速度数据", "tokens": [50750, 2166, 12949, 18032, 6303, 15789, 1546, 31217, 13127, 33188, 26075, 106, 50851], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 36, "seek": 7906, "start": 88.78, "end": 92.12, "text": "都是在真实的MacBook上跑出来的实际", "tokens": [50851, 22796, 3581, 6303, 24726, 1546, 34355, 19203, 5708, 32585, 44561, 1546, 24726, 8842, 227, 51018], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 37, "seek": 7906, "start": 92.12, "end": 92.72, "text": "Benchmark", "tokens": [51018, 21736, 339, 5638, 51048], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 38, "seek": 7906, "start": 92.72, "end": 94.72, "text": "还有那些小图表", "tokens": [51048, 35091, 4184, 13824, 7322, 3919, 122, 17571, 51148], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 39, "seek": 7906, "start": 94.72, "end": 96.48, "text": "就算上下文越来越长", "tokens": [51148, 3111, 19497, 5708, 4438, 17174, 25761, 6912, 25761, 32271, 51236], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 40, "seek": 7906, "start": 96.48, "end": 99.38, "text": "Token Per Second也几乎没怎么掉", "tokens": [51236, 51, 8406, 3026, 5736, 6404, 6336, 254, 2930, 236, 10062, 15282, 29327, 51381], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 41, "seek": 7906, "start": 99.38, "end": 101.2, "text": "然后他会讲到", "tokens": [51381, 26636, 5000, 12949, 39255, 4511, 51472], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 42, "seek": 7906, "start": 101.2, "end": 102.94, "text": "怎么运行比你内存还大的模型", "tokens": [51472, 15282, 3316, 238, 8082, 11706, 2166, 34742, 39781, 14852, 39156, 41908, 39823, 51559], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 43, "seek": 7906, "start": 102.94, "end": 104.34, "text": "快到下面时", "tokens": [51559, 10251, 4511, 47150, 15729, 51629], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 44, "seek": 7906, "start": 104.34, "end": 105.64, "text": "最疯狂的部分来了", "tokens": [51629, 8661, 18140, 107, 18637, 224, 1546, 32174, 34912, 51694], "temperature": 0, "avg_logprob": -0.1199710257101379, "compression_ratio": 1.160883280757098, "no_speech_prob": 1.6753928105961435e-11}, {"id": 45, "seek": 10564, "start": 105.64, "end": 110.48, "text": "把一个模型拆到两台用网线连在一起的笔记本上跑", "tokens": [50365, 16075, 20182, 41908, 39823, 6852, 228, 4511, 36257, 15433, 9254, 16469, 239, 16853, 123, 3316, 252, 3581, 29567, 1546, 5437, 242, 34756, 8802, 5708, 32585, 50607], "temperature": 0, "avg_logprob": -0.15011893518713137, "compression_ratio": 1.1966101694915254, "no_speech_prob": 1.6636164668071274e-11}, {"id": 46, "seek": 10564, "start": 110.48, "end": 111.9, "text": "说实话", "tokens": [50607, 8090, 24726, 21596, 50678], "temperature": 0, "avg_logprob": -0.15011893518713137, "compression_ratio": 1.1966101694915254, "no_speech_prob": 1.6636164668071274e-11}, {"id": 47, "seek": 10564, "start": 111.9, "end": 114.9, "text": "在自己机器上跑模型也不新鲜了", "tokens": [50678, 3581, 17645, 37960, 34386, 5708, 32585, 41908, 39823, 6404, 1960, 12560, 165, 110, 250, 2289, 50828], "temperature": 0, "avg_logprob": -0.15011893518713137, "compression_ratio": 1.1966101694915254, "no_speech_prob": 1.6636164668071274e-11}, {"id": 48, "seek": 10564, "start": 114.9, "end": 115.92, "text": "有Lama CPP", "tokens": [50828, 2412, 43, 2404, 383, 17755, 50879], "temperature": 0, "avg_logprob": -0.15011893518713137, "compression_ratio": 1.1966101694915254, "no_speech_prob": 1.6636164668071274e-11}, {"id": 49, "seek": 10564, "start": 115.92, "end": 118.14, "text": "这个庞大的开源项目", "tokens": [50879, 15368, 6346, 252, 39156, 18937, 47402, 10178, 117, 11386, 50990], "temperature": 0, "avg_logprob": -0.15011893518713137, "compression_ratio": 1.1966101694915254, "no_speech_prob": 1.6636164668071274e-11}, {"id": 50, "seek": 10564, "start": 118.14, "end": 121.26, "text": "整个生态里有一半都是建立在他之上", "tokens": [50990, 27662, 7549, 8244, 3757, 223, 15759, 32241, 30018, 22796, 34157, 24409, 3581, 5000, 9574, 5708, 51146], "temperature": 0, "avg_logprob": -0.15011893518713137, "compression_ratio": 1.1966101694915254, "no_speech_prob": 1.6636164668071274e-11}, {"id": 51, "seek": 10564, "start": 121.26, "end": 124.96000000000001, "text": "Dwarf Star也毫不掩饰地站在他肩膀上", "tokens": [51146, 35, 6925, 69, 5705, 6404, 7256, 104, 1960, 6900, 102, 41520, 108, 10928, 34155, 3581, 5000, 14356, 102, 39708, 222, 5708, 51331], "temperature": 0, "avg_logprob": -0.15011893518713137, "compression_ratio": 1.1966101694915254, "no_speech_prob": 1.6636164668071274e-11}, {"id": 52, "seek": 10564, "start": 124.96000000000001, "end": 128.86, "text": "Alama把本地模型变成了一型命令安装", "tokens": [51331, 9171, 2404, 16075, 8802, 10928, 41908, 39823, 2129, 246, 11336, 2289, 2257, 39823, 25236, 49061, 16206, 8083, 227, 51526], "temperature": 0, "avg_logprob": -0.15011893518713137, "compression_ratio": 1.1966101694915254, "no_speech_prob": 1.6636164668071274e-11}, {"id": 53, "seek": 10564, "start": 128.86, "end": 133.46, "text": "LM Studio又把这一切包装成了友好的桌面应用", "tokens": [51526, 43, 44, 13500, 17047, 16075, 5562, 2257, 23632, 23305, 8083, 227, 11336, 2289, 14765, 20715, 162, 5475, 8833, 44297, 9254, 51756], "temperature": 0, "avg_logprob": -0.15011893518713137, "compression_ratio": 1.1966101694915254, "no_speech_prob": 1.6636164668071274e-11}, {"id": 54, "seek": 13346, "start": 133.46, "end": 136.36, "text": "连不爱用终端的人也能上手", "tokens": [50365, 3316, 252, 1960, 27324, 9254, 10115, 230, 11957, 107, 29979, 6404, 8225, 5708, 11389, 50510], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 55, "seek": 13346, "start": 136.36, "end": 137.96, "text": "但这个项目不一样", "tokens": [50510, 8395, 15368, 10178, 117, 11386, 1960, 2257, 14496, 50590], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 56, "seek": 13346, "start": 137.96, "end": 140.38, "text": "他走的是完全相反的路", "tokens": [50590, 5000, 9575, 24620, 37100, 15106, 22138, 1546, 24658, 50711], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 57, "seek": 13346, "start": 140.38, "end": 142.48000000000002, "text": "他不是什么模型都支持", "tokens": [50711, 5000, 7296, 10440, 41908, 39823, 7182, 35488, 50816], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 58, "seek": 13346, "start": 142.48000000000002, "end": 144.3, "text": "而是只选一个", "tokens": [50816, 11070, 1541, 14003, 2215, 231, 20182, 50907], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 59, "seek": 13346, "start": 144.3, "end": 145.9, "text": "然后把它调到极致", "tokens": [50907, 26636, 42061, 8897, 225, 4511, 7360, 223, 6784, 112, 50987], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 60, "seek": 13346, "start": 145.9, "end": 148.84, "text": "拼命榨出最后一点速度和效果", "tokens": [50987, 6852, 120, 25236, 22989, 101, 7781, 8661, 13547, 49042, 31217, 13127, 12565, 43076, 9319, 51134], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 61, "seek": 13346, "start": 148.84, "end": 150.74, "text": "先说下这个规模", "tokens": [51134, 10108, 8090, 4438, 15368, 6758, 226, 41908, 51229], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 62, "seek": 13346, "start": 150.74, "end": 153.3, "text": "已经有18500多个Star了", "tokens": [51229, 49161, 2412, 6494, 7526, 6392, 7549, 24659, 2289, 51357], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 63, "seek": 13346, "start": 153.3, "end": 155.72, "text": "而且是MIT许可证", "tokens": [51357, 22942, 1541, 44, 3927, 7422, 116, 4429, 5233, 223, 51478], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 64, "seek": 13346, "start": 155.72, "end": 157.98000000000002, "text": "所以你基本可以随便拿去用", "tokens": [51478, 7239, 2166, 37946, 6723, 10673, 237, 27364, 24351, 6734, 9254, 51591], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 65, "seek": 13346, "start": 157.98000000000002, "end": 160.82, "text": "整个项目几乎完全用C写成", "tokens": [51591, 27662, 7549, 10178, 117, 11386, 6336, 254, 2930, 236, 37100, 9254, 34, 5676, 247, 11336, 51733], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 66, "seek": 13346, "start": 160.82, "end": 161.82, "text": "真想跑得顺", "tokens": [51733, 6303, 7093, 32585, 5916, 10178, 118, 51783], "temperature": 0, "avg_logprob": -0.12841945366571414, "compression_ratio": 1.1872909698996656, "no_speech_prob": 1.590907440507383e-11}, {"id": 67, "seek": 16182, "start": 161.82, "end": 165.42, "text": "还是得有一台差不多96G内存的Mac", "tokens": [50365, 45726, 5916, 32241, 15433, 37876, 22962, 38, 34742, 39781, 1546, 34355, 50545], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 68, "seek": 16182, "start": 165.42, "end": 168.57999999999998, "text": "当然内存更少也能流逝运行", "tokens": [50545, 40486, 34742, 39781, 19002, 15686, 6404, 8225, 27854, 2215, 251, 3316, 238, 8082, 50703], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 69, "seek": 16182, "start": 168.57999999999998, "end": 171.73999999999998, "text": "好机制上它到底是怎么工作的", "tokens": [50703, 2131, 37960, 25491, 5708, 11284, 33883, 1541, 15282, 41315, 1546, 50861], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 70, "seek": 16182, "start": 171.73999999999998, "end": 174.14, "text": "核心是在现代机器上", "tokens": [50861, 8987, 116, 7945, 1541, 3581, 20204, 19105, 37960, 34386, 5708, 50981], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 71, "seek": 16182, "start": 174.14, "end": 177.85999999999999, "text": "你的模型和缓存不一定非得放在RAM里", "tokens": [50981, 18961, 41908, 39823, 12565, 38109, 241, 39781, 1960, 24272, 12107, 5916, 12744, 3581, 49, 2865, 15759, 51167], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 72, "seek": 16182, "start": 177.85999999999999, "end": 180.24, "text": "也可以放在SSD上", "tokens": [51167, 47935, 12744, 3581, 21929, 35, 5708, 51286], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 73, "seek": 16182, "start": 180.24, "end": 181.72, "text": "这个思路一变", "tokens": [51286, 15368, 8870, 24658, 2257, 2129, 246, 51360], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 74, "seek": 16182, "start": 181.72, "end": 183.66, "text": "事情就完全不一样了", "tokens": [51360, 24675, 3111, 37100, 1960, 2257, 14496, 2289, 51457], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 75, "seek": 16182, "start": 183.66, "end": 185.34, "text": "路由专家层", "tokens": [51457, 24658, 23786, 940, 241, 5155, 9636, 224, 51541], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 76, "seek": 16182, "start": 185.34, "end": 187.88, "text": "也就是这类Mixture Model里", "tokens": [51541, 6404, 5620, 5562, 22113, 119, 44, 72, 8890, 17105, 15759, 51668], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 77, "seek": 16182, "start": 187.88, "end": 189.0, "text": "占大头的部分", "tokens": [51668, 5322, 254, 3582, 39862, 1546, 32174, 51724], "temperature": 0, "avg_logprob": -0.16846104273720394, "compression_ratio": 1.158273381294964, "no_speech_prob": 1.5147742435384082e-11}, {"id": 78, "seek": 18900, "start": 189.0, "end": 191.72, "text": "会在某个Token需要它的时候", "tokens": [50365, 12949, 3581, 17238, 238, 7549, 51, 8406, 35748, 45224, 29111, 50501], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 79, "seek": 18900, "start": 191.72, "end": 193.16, "text": "才从磁盘流逝加赛", "tokens": [50501, 18888, 35630, 163, 96, 223, 5419, 246, 27854, 2215, 251, 9990, 5266, 249, 50573], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 80, "seek": 18900, "start": 193.16, "end": 194.46, "text": "这样一来", "tokens": [50573, 21209, 2257, 6912, 50638], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 81, "seek": 18900, "start": 194.46, "end": 197.22, "text": "内存就不再是一睹你一头撞上的硬墙", "tokens": [50638, 34742, 39781, 3111, 1960, 8623, 1541, 2257, 14221, 117, 2166, 2257, 39862, 20559, 252, 5708, 1546, 38114, 105, 24228, 247, 50776], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 82, "seek": 18900, "start": 197.22, "end": 199.48, "text": "而更像是加速用的杠杆", "tokens": [50776, 11070, 19002, 12760, 1541, 9990, 31217, 9254, 1546, 4422, 254, 4422, 228, 50889], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 83, "seek": 18900, "start": 199.48, "end": 201.44, "text": "为了让它能跑起来", "tokens": [50889, 13992, 2289, 33650, 11284, 8225, 32585, 9147, 6912, 50987], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 84, "seek": 18900, "start": 201.44, "end": 204.02, "text": "它用了比较激进的2bit量化", "tokens": [50987, 11284, 9254, 2289, 11706, 9830, 225, 26373, 222, 36700, 1546, 17, 5260, 26748, 23756, 51116], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 85, "seek": 18900, "start": 204.02, "end": 206.72, "text": "但这是有意做成不对称的", "tokens": [51116, 8395, 27455, 2412, 9042, 10907, 11336, 41639, 8204, 108, 1546, 51251], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 86, "seek": 18900, "start": 206.72, "end": 208.56, "text": "它只压缩Exper权重", "tokens": [51251, 11284, 14003, 5014, 233, 38109, 102, 11149, 610, 4422, 225, 12624, 51343], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 87, "seek": 18900, "start": 208.56, "end": 210.82, "text": "Attention和共享层", "tokens": [51343, 38151, 1251, 12565, 15408, 27654, 9636, 224, 51456], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 88, "seek": 18900, "start": 210.82, "end": 212.9, "text": "则尽量保持接近满质量", "tokens": [51456, 2437, 247, 1530, 121, 26748, 24302, 17694, 14468, 17463, 15868, 94, 18464, 101, 26748, 51560], "temperature": 0, "avg_logprob": -0.12966576667681132, "compression_ratio": 1.1134751773049645, "no_speech_prob": 1.6746607572892813e-11}, {"id": 89, "seek": 21290, "start": 212.9, "end": 215.9, "text": "所以它还能保持足够稳定", "tokens": [50365, 7239, 11284, 14852, 8225, 24302, 17694, 37236, 1787, 253, 10415, 111, 12088, 50515], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 90, "seek": 21290, "start": 215.9, "end": 217.76000000000002, "text": "值得拿来做真实任务", "tokens": [50515, 40242, 5916, 24351, 6912, 10907, 6303, 24726, 26443, 5087, 94, 50608], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 91, "seek": 21290, "start": 217.76000000000002, "end": 219.84, "text": "而结果确实有效", "tokens": [50608, 11070, 45641, 9319, 38114, 106, 24726, 2412, 43076, 50712], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 92, "seek": 21290, "start": 219.84, "end": 221.82, "text": "在M3MAX笔记本上", "tokens": [50712, 3581, 44, 18, 9998, 55, 5437, 242, 34756, 8802, 5708, 50811], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 93, "seek": 21290, "start": 221.82, "end": 224.76, "text": "速度大约是每秒27个Token", "tokens": [50811, 31217, 13127, 3582, 16853, 99, 1541, 23664, 20956, 10076, 7549, 51, 8406, 50958], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 94, "seek": 21290, "start": 224.76, "end": 226.26, "text": "已经完全够用了", "tokens": [50958, 49161, 37100, 1787, 253, 9254, 2289, 51033], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 95, "seek": 21290, "start": 226.26, "end": 228.04000000000002, "text": "想被推荐吗", "tokens": [51033, 7093, 23238, 33597, 31409, 238, 14769, 51122], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 96, "seek": 21290, "start": 228.04000000000002, "end": 230.72, "text": "把你的开源仓库提交到Indie Hacker", "tokens": [51122, 16075, 18961, 18937, 47402, 1550, 241, 6346, 241, 20949, 28455, 4511, 21790, 414, 389, 23599, 51256], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 97, "seek": 21290, "start": 230.72, "end": 232.20000000000002, "text": "New Supply", "tokens": [51256, 18278, 9391, 356, 51330], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 98, "seek": 21290, "start": 232.20000000000002, "end": 234.9, "text": "就能让成千上万的开发者看到你", "tokens": [51330, 3111, 8225, 33650, 11336, 20787, 5708, 23570, 1546, 18937, 28926, 12444, 18032, 2166, 51465], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 99, "seek": 21290, "start": 234.9, "end": 236.24, "text": "说实话", "tokens": [51465, 8090, 24726, 21596, 51532], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 100, "seek": 21290, "start": 236.24, "end": 238.34, "text": "到这里它就不再只是个技术演示了", "tokens": [51532, 4511, 35102, 11284, 3111, 1960, 8623, 36859, 7549, 32502, 1474, 107, 31382, 25696, 2289, 51637], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 101, "seek": 21290, "start": 238.34, "end": 240.54000000000002, "text": "而是开始像个真正的产品了", "tokens": [51637, 11070, 1541, 45213, 12760, 7549, 6303, 15789, 1546, 1369, 100, 30246, 2289, 51747], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 102, "seek": 21290, "start": 240.54000000000002, "end": 242.20000000000002, "text": "它内置了一个编码Agent", "tokens": [51747, 11284, 34742, 34719, 2289, 20182, 38109, 244, 23230, 223, 32, 6930, 51830], "temperature": 0, "avg_logprob": -0.14214998881022137, "compression_ratio": 1.137313432835821, "no_speech_prob": 1.6820024539843104e-11}, {"id": 103, "seek": 24220, "start": 242.2, "end": 243.92, "text": "有个很巧的设计师", "tokens": [50365, 2412, 7549, 4563, 43605, 1546, 7422, 122, 7422, 94, 29186, 50451], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 104, "seek": 24220, "start": 243.92, "end": 246.06, "text": "这个绘画不会存到什么数据库里", "tokens": [50451, 15368, 10115, 246, 27126, 47928, 39781, 4511, 10440, 33188, 26075, 106, 6346, 241, 15759, 50558], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 105, "seek": 24220, "start": 246.06, "end": 248.76, "text": "这个Session本质上就是把KVCache", "tokens": [50558, 15368, 50, 4311, 8802, 18464, 101, 5708, 5620, 16075, 42, 53, 34, 6000, 50693], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 106, "seek": 24220, "start": 248.76, "end": 249.98, "text": "放在你的硬盘上", "tokens": [50693, 12744, 3581, 18961, 38114, 105, 5419, 246, 5708, 50754], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 107, "seek": 24220, "start": 249.98, "end": 253.0, "text": "这样切换以保存的对话就会非常快", "tokens": [50754, 21209, 23632, 26075, 95, 3588, 24302, 39781, 1546, 8713, 21596, 3111, 12949, 14392, 10251, 50905], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 108, "seek": 24220, "start": 253.0, "end": 254.64, "text": "它还内置了一个", "tokens": [50905, 11284, 14852, 34742, 34719, 2289, 20182, 50987], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 109, "seek": 24220, "start": 254.64, "end": 257.8, "text": "同时支持OpenLayer和Enthropy API的Server", "tokens": [50987, 13089, 15729, 35488, 45569, 43, 11167, 12565, 16257, 14222, 88, 9362, 1546, 31859, 331, 51145], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 110, "seek": 24220, "start": 257.8, "end": 261.94, "text": "这意味着你可以直接把Codex或Cloud Code", "tokens": [51145, 5562, 9042, 17268, 20708, 42766, 43297, 16075, 34, 1429, 87, 19780, 32787, 15549, 51352], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 111, "seek": 24220, "start": 261.94, "end": 263.82, "text": "指向你自己的笔记本电脑", "tokens": [51352, 25922, 24282, 2166, 17645, 1546, 5437, 242, 34756, 8802, 42182, 27067, 239, 51446], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 112, "seek": 24220, "start": 263.82, "end": 266.38, "text": "在你自己托管的模型上工作", "tokens": [51446, 3581, 2166, 17645, 3416, 246, 23131, 1546, 41908, 39823, 5708, 41315, 51574], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 113, "seek": 24220, "start": 266.38, "end": 268.62, "text": "它甚至加了一个PowerDial", "tokens": [51574, 11284, 17556, 20844, 9990, 2289, 20182, 46057, 35, 831, 51686], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 114, "seek": 24220, "start": 268.62, "end": 271.38, "text": "可以在你更在意机器两款安静", "tokens": [51686, 6723, 3581, 2166, 19002, 3581, 9042, 37960, 34386, 36257, 48798, 16206, 5363, 247, 51824], "temperature": 0, "avg_logprob": -0.1359928341730972, "compression_ratio": 1.192090395480226, "no_speech_prob": 1.4275108005390447e-11}, {"id": 115, "seek": 27138, "start": 271.38, "end": 274.92, "text": "而不是榨干每一个Token的时候去限制GPU", "tokens": [50365, 11070, 7296, 22989, 101, 26111, 23664, 20182, 51, 8406, 49873, 6734, 43446, 25491, 38, 8115, 50542], "temperature": 0, "avg_logprob": -0.12424365211935605, "compression_ratio": 1.0893617021276596, "no_speech_prob": 1.604941526900383e-11}, {"id": 116, "seek": 27138, "start": 274.92, "end": 277.65999999999997, "text": "所有这些都在指向同一个想法", "tokens": [50542, 39300, 5562, 13824, 7182, 3581, 25922, 24282, 13089, 20182, 7093, 11148, 50679], "temperature": 0, "avg_logprob": -0.12424365211935605, "compression_ratio": 1.0893617021276596, "no_speech_prob": 1.604941526900383e-11}, {"id": 117, "seek": 27138, "start": 277.65999999999997, "end": 280.76, "text": "本地模型应该像一个做完整的产品", "tokens": [50679, 8802, 10928, 41908, 39823, 44297, 44646, 12760, 20182, 10907, 14128, 27662, 1546, 1369, 100, 30246, 50834], "temperature": 0, "avg_logprob": -0.12424365211935605, "compression_ratio": 1.0893617021276596, "no_speech_prob": 1.604941526900383e-11}, {"id": 118, "seek": 27138, "start": 280.76, "end": 282.06, "text": "而不是一堆零件", "tokens": [50834, 11070, 7296, 2257, 10726, 228, 6306, 114, 20485, 50899], "temperature": 0, "avg_logprob": -0.12424365211935605, "compression_ratio": 1.0893617021276596, "no_speech_prob": 1.604941526900383e-11}, {"id": 119, "seek": 27138, "start": 282.06, "end": 283.18, "text": "说到底", "tokens": [50899, 8090, 33883, 50955], "temperature": 0, "avg_logprob": -0.12424365211935605, "compression_ratio": 1.0893617021276596, "no_speech_prob": 1.604941526900383e-11}, {"id": 120, "seek": 27138, "start": 283.18, "end": 284.86, "text": "它的底层其实非常老派", "tokens": [50955, 45224, 22816, 9636, 224, 41646, 14392, 10439, 12136, 122, 51039], "temperature": 0, "avg_logprob": -0.12424365211935605, "compression_ratio": 1.0893617021276596, "no_speech_prob": 1.604941526900383e-11}, {"id": 121, "seek": 27138, "start": 284.86, "end": 286.62, "text": "核心部分是纯C", "tokens": [51039, 8987, 116, 7945, 32174, 1541, 16853, 107, 34, 51127], "temperature": 0, "avg_logprob": -0.12424365211935605, "compression_ratio": 1.0893617021276596, "no_speech_prob": 1.604941526900383e-11}, {"id": 122, "seek": 27138, "start": 286.62, "end": 289.94, "text": "再加上一些手写的Cuda和Metal Shader", "tokens": [51127, 8623, 9990, 5708, 38515, 11389, 5676, 247, 1546, 34, 11152, 12565, 44, 302, 304, 1160, 8312, 51293], "temperature": 0, "avg_logprob": -0.12424365211935605, "compression_ratio": 1.0893617021276596, "no_speech_prob": 1.604941526900383e-11}, {"id": 123, "seek": 28994, "start": 289.94, "end": 293.54, "text": "真正负责GPU运算的几乎就只有这些", "tokens": [50365, 6303, 15789, 18464, 253, 18464, 96, 38, 8115, 3316, 238, 19497, 1546, 6336, 254, 2930, 236, 3111, 35244, 5562, 13824, 50545], "temperature": 0, "avg_logprob": -0.14709230858509934, "compression_ratio": 1.0898876404494382, "no_speech_prob": 1.3775756577127929e-11}, {"id": 124, "seek": 28994, "start": 293.54, "end": 296.1, "text": "它明确没有链接GGML", "tokens": [50545, 11284, 11100, 38114, 106, 17944, 165, 241, 122, 14468, 38, 38, 12683, 50673], "temperature": 0, "avg_logprob": -0.14709230858509934, "compression_ratio": 1.0898876404494382, "no_speech_prob": 1.3775756577127929e-11}, {"id": 125, "seek": 28994, "start": 296.1, "end": 298.06, "text": "也就是LamaCPP背后的那个库", "tokens": [50673, 6404, 5620, 43, 2404, 34, 17755, 46329, 13547, 1546, 43083, 6346, 241, 50771], "temperature": 0, "avg_logprob": -0.14709230858509934, "compression_ratio": 1.0898876404494382, "no_speech_prob": 1.3775756577127929e-11}, {"id": 126, "seek": 28994, "start": 298.06, "end": 301.71999999999997, "text": "但它确实适配了其中一些格式和Kernel", "tokens": [50771, 8395, 11284, 38114, 106, 24726, 2215, 224, 38846, 2289, 9572, 5975, 38515, 30921, 27584, 12565, 42, 1248, 338, 50954], "temperature": 0, "avg_logprob": -0.14709230858509934, "compression_ratio": 1.0898876404494382, "no_speech_prob": 1.3775756577127929e-11}, {"id": 127, "seek": 28994, "start": 301.71999999999997, "end": 302.94, "text": "而且出于尊重", "tokens": [50954, 22942, 7781, 37732, 1530, 232, 12624, 51015], "temperature": 0, "avg_logprob": -0.14709230858509934, "compression_ratio": 1.0898876404494382, "no_speech_prob": 1.3775756577127929e-11}, {"id": 128, "seek": 28994, "start": 302.94, "end": 306.48, "text": "把它们的版权信息直接保留了下来", "tokens": [51015, 42061, 9497, 1546, 42096, 4422, 225, 17665, 26460, 43297, 24302, 24456, 2289, 4438, 6912, 51192], "temperature": 0, "avg_logprob": -0.14709230858509934, "compression_ratio": 1.0898876404494382, "no_speech_prob": 1.3775756577127929e-11}, {"id": 129, "seek": 28994, "start": 306.48, "end": 310.86, "text": "每次构建都会拿DivC在不同上下文长度下的官方输出做检查", "tokens": [51192, 23664, 9487, 7360, 226, 34157, 7182, 12949, 24351, 35, 592, 34, 3581, 47123, 5708, 4438, 17174, 32271, 13127, 4438, 1546, 31929, 9249, 9830, 241, 7781, 10907, 16407, 222, 42623, 51411], "temperature": 0, "avg_logprob": -0.14709230858509934, "compression_ratio": 1.0898876404494382, "no_speech_prob": 1.3775756577127929e-11}, {"id": 130, "seek": 31086, "start": 310.86, "end": 314.72, "text": "所以那种悄悄把质量搞坏的改动会在发布前被发现", "tokens": [50365, 7239, 4184, 39810, 14696, 226, 14696, 226, 16075, 18464, 101, 26748, 41816, 14872, 237, 1546, 34490, 34961, 12949, 3581, 28926, 34688, 8945, 23238, 28926, 20204, 50558], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 131, "seek": 31086, "start": 314.72, "end": 317.6, "text": "它还能跑在Apple Metal和NVIDIA显卡上", "tokens": [50558, 11284, 14852, 8225, 32585, 3581, 9132, 306, 23488, 12565, 45, 3958, 6914, 1431, 122, 32681, 5708, 50702], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 132, "seek": 31086, "start": 317.6, "end": 319.94, "text": "连小小的DGX2也支持", "tokens": [50702, 3316, 252, 7322, 7322, 1546, 35, 38, 55, 17, 6404, 35488, 50819], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 133, "seek": 31086, "start": 319.94, "end": 323.72, "text": "它甚至还能处理AMD的统一内存机器", "tokens": [50819, 11284, 17556, 20844, 14852, 8225, 1787, 226, 13876, 2865, 35, 1546, 10115, 253, 2257, 34742, 39781, 37960, 34386, 51008], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 134, "seek": 31086, "start": 323.72, "end": 326.1, "text": "启动它出人意料的简单", "tokens": [51008, 2392, 107, 34961, 11284, 7781, 4035, 9042, 33404, 1546, 11249, 222, 47446, 51127], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 135, "seek": 31086, "start": 326.1, "end": 327.3, "text": "甚至有点无聊", "tokens": [51127, 17556, 20844, 2412, 12579, 31364, 40096, 51187], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 136, "seek": 31086, "start": 327.3, "end": 328.36, "text": "但这是好事", "tokens": [51187, 8395, 27455, 2131, 6973, 51240], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 137, "seek": 31086, "start": 328.36, "end": 330.36, "text": "你先运行下载脚本", "tokens": [51240, 2166, 10108, 3316, 238, 8082, 4438, 17819, 121, 27067, 248, 8802, 51340], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 138, "seek": 31086, "start": 330.36, "end": 332.8, "text": "把适配你机器的2bit模型拉下来", "tokens": [51340, 16075, 2215, 224, 38846, 2166, 37960, 34386, 1546, 17, 5260, 41908, 39823, 29717, 4438, 6912, 51462], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 139, "seek": 31086, "start": 332.8, "end": 335.96000000000004, "text": "然后在Mac上用Mac构建Metal版本", "tokens": [51462, 26636, 3581, 34355, 5708, 9254, 34355, 7360, 226, 34157, 44, 302, 304, 42096, 8802, 51620], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 140, "seek": 31086, "start": 335.96000000000004, "end": 337.12, "text": "再把它启动起来", "tokens": [51620, 8623, 42061, 2392, 107, 34961, 9147, 6912, 51678], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 141, "seek": 31086, "start": 337.12, "end": 340.22, "text": "你还可以选择开启SSD Streaming", "tokens": [51678, 2166, 14852, 6723, 2215, 231, 6852, 102, 18937, 2392, 107, 21929, 35, 24904, 278, 51833], "temperature": 0, "avg_logprob": -0.10359456565942657, "compression_ratio": 1.1783625730994152, "no_speech_prob": 1.3476645147059152e-11}, {"id": 142, "seek": 34086, "start": 340.86, "end": 343.28000000000003, "text": "这样就能用比实际内存更大的模型", "tokens": [50365, 21209, 3111, 8225, 9254, 11706, 24726, 8842, 227, 34742, 39781, 19002, 39156, 41908, 39823, 50486], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 143, "seek": 34086, "start": 343.28000000000003, "end": 344.02000000000004, "text": "之后", "tokens": [50486, 9574, 13547, 50523], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 144, "seek": 34086, "start": 344.02000000000004, "end": 347.12, "text": "你就是在和自己硬件上的模型对话", "tokens": [50523, 2166, 5620, 3581, 12565, 17645, 38114, 105, 20485, 5708, 1546, 41908, 39823, 8713, 21596, 50678], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 145, "seek": 34086, "start": 347.12, "end": 350.04, "text": "输入的内容也不会离开这台机器", "tokens": [50678, 9830, 241, 14028, 1546, 34742, 25750, 6404, 47928, 18614, 119, 18937, 5562, 15433, 37960, 34386, 50824], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 146, "seek": 34086, "start": 350.04, "end": 352.68, "text": "然后还有一个最近才上线", "tokens": [50824, 26636, 35091, 20182, 46777, 18888, 5708, 16853, 123, 50956], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 147, "seek": 34086, "start": 352.68, "end": 354.90000000000003, "text": "让大家都有点惊讶的功能", "tokens": [50956, 33650, 46525, 2412, 12579, 4799, 232, 7422, 114, 1546, 32311, 8225, 51067], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 148, "seek": 34086, "start": 354.90000000000003, "end": 356.7, "text": "Distributed Mode", "tokens": [51067, 35, 468, 2024, 4866, 20500, 51157], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 149, "seek": 34086, "start": 356.7, "end": 360.06, "text": "它会把一个模型拆到多台机器上", "tokens": [51157, 11284, 12949, 16075, 20182, 41908, 39823, 6852, 228, 4511, 6392, 15433, 37960, 34386, 5708, 51325], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 150, "seek": 34086, "start": 360.06, "end": 362.26, "text": "每台分一部分layers", "tokens": [51325, 23664, 15433, 6627, 2257, 32174, 8376, 433, 51435], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 151, "seek": 34086, "start": 362.26, "end": 365.88, "text": "这样一个大到任何单机都跑不动的模型", "tokens": [51435, 21209, 20182, 3582, 4511, 26443, 11906, 47446, 37960, 7182, 32585, 1960, 34961, 1546, 41908, 39823, 51616], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 152, "seek": 34086, "start": 365.88, "end": 367.40000000000003, "text": "也照样能运行", "tokens": [51616, 6404, 32150, 14496, 8225, 3316, 238, 8082, 51692], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 153, "seek": 34086, "start": 367.40000000000003, "end": 369.52000000000004, "text": "最典型的例子", "tokens": [51692, 8661, 2347, 116, 39823, 1546, 17797, 7626, 51798], "temperature": 0, "avg_logprob": -0.1162262225973195, "compression_ratio": 1.2516778523489933, "no_speech_prob": 1.4864420458393646e-11}, {"id": 154, "seek": 36952, "start": 369.52, "end": 373.94, "text": "是两台128G的MacBook通过Thunderbolt连接", "tokens": [50365, 1541, 36257, 15433, 4762, 23, 38, 1546, 34355, 19203, 19550, 16866, 2434, 6617, 39477, 3316, 252, 14468, 50586], "temperature": 0, "avg_logprob": -0.12532633404399074, "compression_ratio": 1.1393728222996515, "no_speech_prob": 1.861694652605106e-11}, {"id": 155, "seek": 36952, "start": 373.94, "end": 377.64, "text": "在它们之间运行完整的4bit模型", "tokens": [50586, 3581, 11284, 9497, 9574, 31685, 3316, 238, 8082, 14128, 27662, 1546, 19, 5260, 41908, 39823, 50771], "temperature": 0, "avg_logprob": -0.12532633404399074, "compression_ratio": 1.1393728222996515, "no_speech_prob": 1.861694652605106e-11}, {"id": 156, "seek": 36952, "start": 377.64, "end": 379.24, "text": "更惊人的是", "tokens": [50771, 19002, 4799, 232, 4035, 24620, 50851], "temperature": 0, "avg_logprob": -0.12532633404399074, "compression_ratio": 1.1393728222996515, "no_speech_prob": 1.861694652605106e-11}, {"id": 157, "seek": 36952, "start": 379.24, "end": 382.71999999999997, "text": "这一切都是在最近几周的提交里才出现的", "tokens": [50851, 5562, 2257, 23632, 22796, 3581, 46777, 6336, 254, 37390, 1546, 20949, 28455, 15759, 18888, 7781, 20204, 1546, 51025], "temperature": 0, "avg_logprob": -0.12532633404399074, "compression_ratio": 1.1393728222996515, "no_speech_prob": 1.861694652605106e-11}, {"id": 158, "seek": 36952, "start": 382.71999999999997, "end": 384.97999999999996, "text": "所以这对你意味着什么", "tokens": [51025, 7239, 5562, 8713, 2166, 9042, 17268, 20708, 10440, 51138], "temperature": 0, "avg_logprob": -0.12532633404399074, "compression_ratio": 1.1393728222996515, "no_speech_prob": 1.861694652605106e-11}, {"id": 159, "seek": 36952, "start": 384.97999999999996, "end": 386.26, "text": "现在还很早", "tokens": [51138, 25040, 14852, 4563, 21176, 51202], "temperature": 0, "avg_logprob": -0.12532633404399074, "compression_ratio": 1.1393728222996515, "no_speech_prob": 1.861694652605106e-11}, {"id": 160, "seek": 36952, "start": 386.26, "end": 388.71999999999997, "text": "作者也首先承认这还是Beta", "tokens": [51202, 11914, 12444, 6404, 36490, 3416, 123, 7422, 97, 5562, 45726, 33, 7664, 51325], "temperature": 0, "avg_logprob": -0.12532633404399074, "compression_ratio": 1.1393728222996515, "no_speech_prob": 1.861694652605106e-11}, {"id": 161, "seek": 36952, "start": 388.71999999999997, "end": 391.38, "text": "已经有几百个未解决的问题", "tokens": [51325, 49161, 2412, 6336, 254, 31906, 7549, 29954, 17278, 5676, 111, 1546, 34069, 51458], "temperature": 0, "avg_logprob": -0.12532633404399074, "compression_ratio": 1.1393728222996515, "no_speech_prob": 1.861694652605106e-11}, {"id": 162, "seek": 36952, "start": 391.38, "end": 395.35999999999996, "text": "而且它只会运行它发布的那些特定模型文件", "tokens": [51458, 22942, 11284, 14003, 12949, 3316, 238, 8082, 11284, 28926, 34688, 1546, 4184, 13824, 17682, 12088, 41908, 39823, 17174, 20485, 51657], "temperature": 0, "avg_logprob": -0.12532633404399074, "compression_ratio": 1.1393728222996515, "no_speech_prob": 1.861694652605106e-11}, {"id": 163, "seek": 39536, "start": 395.36, "end": 398.26, "text": "所以它是把利器不是瑞士军刀", "tokens": [50365, 7239, 11284, 1541, 16075, 23700, 34386, 7296, 37955, 252, 30337, 5676, 249, 2437, 222, 50510], "temperature": 0, "avg_logprob": -0.14962319070978683, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.6372830174971043e-11}, {"id": 164, "seek": 39536, "start": 398.26, "end": 402.36, "text": "你还得有一台相当强的Mac才能获得好的体验", "tokens": [50510, 2166, 14852, 5916, 32241, 15433, 15106, 16233, 5702, 118, 1546, 34355, 18888, 8225, 31127, 115, 5916, 20715, 29485, 49657, 234, 50715], "temperature": 0, "avg_logprob": -0.14962319070978683, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.6372830174971043e-11}, {"id": 165, "seek": 39536, "start": 402.36, "end": 404.28000000000003, "text": "也有些人会介意", "tokens": [50715, 6404, 2412, 13824, 4035, 12949, 30312, 9042, 50811], "temperature": 0, "avg_logprob": -0.14962319070978683, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.6372830174971043e-11}, {"id": 166, "seek": 39536, "start": 404.28000000000003, "end": 407.96000000000004, "text": "这里面很多代码都是在AI参与下写出来的", "tokens": [50811, 35102, 8833, 20778, 19105, 23230, 223, 22796, 3581, 48698, 2129, 224, 940, 236, 4438, 5676, 247, 44561, 1546, 50995], "temperature": 0, "avg_logprob": -0.14962319070978683, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.6372830174971043e-11}, {"id": 167, "seek": 39536, "start": 407.96000000000004, "end": 409.02000000000004, "text": "但往后看", "tokens": [50995, 8395, 29510, 13547, 4200, 51048], "temperature": 0, "avg_logprob": -0.14962319070978683, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.6372830174971043e-11}, {"id": 168, "seek": 39536, "start": 409.02000000000004, "end": 412.16, "text": "这背后的压注其实挺让人兴奋的", "tokens": [51048, 5562, 46329, 13547, 1546, 5014, 233, 26432, 41646, 41046, 33650, 4035, 2347, 112, 1881, 233, 1546, 51205], "temperature": 0, "avg_logprob": -0.14962319070978683, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.6372830174971043e-11}, {"id": 169, "seek": 39536, "start": 412.16, "end": 412.74, "text": "它在说", "tokens": [51205, 11284, 3581, 8090, 51234], "temperature": 0, "avg_logprob": -0.14962319070978683, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.6372830174971043e-11}, {"id": 170, "seek": 39536, "start": 412.74, "end": 416.3, "text": "这么强大的模型完全跑在你自己的硬件上", "tokens": [51234, 31889, 5702, 118, 39156, 41908, 39823, 37100, 32585, 3581, 2166, 17645, 1546, 38114, 105, 20485, 5708, 51412], "temperature": 0, "avg_logprob": -0.14962319070978683, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.6372830174971043e-11}, {"id": 171, "seek": 39536, "start": 416.3, "end": 419.02000000000004, "text": "而且不会把任何东西回传出去", "tokens": [51412, 22942, 47928, 16075, 26443, 11906, 38409, 16220, 8350, 7384, 254, 48973, 51548], "temperature": 0, "avg_logprob": -0.14962319070978683, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.6372830174971043e-11}, {"id": 172, "seek": 41902, "start": 419.02, "end": 420.71999999999997, "text": "这不是以后才会有的事", "tokens": [50365, 5562, 7296, 3588, 13547, 18888, 12949, 2412, 1546, 6973, 50450], "temperature": 0, "avg_logprob": -0.11003329611232138, "compression_ratio": 1.2560975609756098, "no_speech_prob": 1.473910056504213e-11}, {"id": 173, "seek": 41902, "start": 420.71999999999997, "end": 422.96, "text": "而是现在就能做到的事", "tokens": [50450, 11070, 1541, 25040, 3111, 8225, 10907, 4511, 1546, 6973, 50562], "temperature": 0, "avg_logprob": -0.11003329611232138, "compression_ratio": 1.2560975609756098, "no_speech_prob": 1.473910056504213e-11}, {"id": 174, "seek": 41902, "start": 422.96, "end": 426.38, "text": "如果你是那种想把整套技术站都摆在桌面上", "tokens": [50562, 13119, 32526, 4184, 39810, 7093, 16075, 27662, 1881, 245, 32502, 1474, 107, 34155, 7182, 34783, 228, 3581, 162, 5475, 8833, 5708, 50733], "temperature": 0, "avg_logprob": -0.11003329611232138, "compression_ratio": 1.2560975609756098, "no_speech_prob": 1.473910056504213e-11}, {"id": 175, "seek": 41902, "start": 426.38, "end": 428.78, "text": "而不是放在别人云端里的开发者", "tokens": [50733, 11070, 7296, 12744, 3581, 18453, 4035, 1369, 239, 11957, 107, 15759, 1546, 18937, 28926, 12444, 50853], "temperature": 0, "avg_logprob": -0.11003329611232138, "compression_ratio": 1.2560975609756098, "no_speech_prob": 1.473910056504213e-11}, {"id": 176, "seek": 41902, "start": 428.78, "end": 431.21999999999997, "text": "这大概就是目前最清楚的信号", "tokens": [50853, 5562, 32044, 5620, 39004, 8661, 38433, 1546, 17665, 26987, 50975], "temperature": 0, "avg_logprob": -0.11003329611232138, "compression_ratio": 1.2560975609756098, "no_speech_prob": 1.473910056504213e-11}, {"id": 177, "seek": 41902, "start": 431.21999999999997, "end": 432.53999999999996, "text": "说明局面正在变化", "tokens": [50975, 8090, 11100, 34703, 8833, 15789, 3581, 2129, 246, 23756, 51041], "temperature": 0, "avg_logprob": -0.11003329611232138, "compression_ratio": 1.2560975609756098, "no_speech_prob": 1.473910056504213e-11}, {"id": 178, "seek": 41902, "start": 432.53999999999996, "end": 435.56, "text": "再想到这位作者下一次他发出来的东西", "tokens": [51041, 8623, 7093, 4511, 5562, 11160, 11914, 12444, 4438, 27505, 5000, 28926, 44561, 1546, 38409, 16220, 51192], "temperature": 0, "avg_logprob": -0.11003329611232138, "compression_ratio": 1.2560975609756098, "no_speech_prob": 1.473910056504213e-11}, {"id": 179, "seek": 41902, "start": 435.56, "end": 437.29999999999995, "text": "肯定也会很有意思", "tokens": [51192, 14356, 107, 12088, 6404, 12949, 49295, 16697, 51279], "temperature": 0, "avg_logprob": -0.11003329611232138, "compression_ratio": 1.2560975609756098, "no_speech_prob": 1.473910056504213e-11}, {"id": 180, "seek": 41902, "start": 437.29999999999995, "end": 438.82, "text": "值得一看", "tokens": [51279, 40242, 5916, 2257, 4200, 51355], "temperature": 0, "avg_logprob": -0.11003329611232138, "compression_ratio": 1.2560975609756098, "no_speech_prob": 1.473910056504213e-11}], "language": "zh"}