{ "text": "這是你可以在消費級硬件上運行的最好的本地模型。而且就其规模而言,它真的非常出色。在本視頻中,我將向你展示如何將其與DeepSeekHarness這樣強大的編碼代理結合,來構建強大的代理系統。而且它還有一些非常有趣的隱藏能力。例如,它在圖像理解和繪製邊界框方面非常出色。所以在本視頻中,我的目標是向你展示如何設置它。\n如果你想使用这个模型\n应该使用哪些不同的配置\n以及许多人在推理时常犯的错误\n首先我们来看几个不同的基准测试\n因为这确实能展示这个模型的强大之处\n这是人工智能代理指数\n目前是51\n实际上仅落后于KIMIK3\nKIMIK3是一个2.8万亿参数的模型\n最重要的是性能在真实世界测试中依然保持\n稍后我会在视频中展示几个例子\n现在来对比一下\n它的智能和性能非常接近最高设置下的Cloud 4.8\n而那是几个月前的顶尖模型\n有人可能会说这可能是基准测试的结果\n但真实世界的性能\n实际上非常非常出色\n我将向你展示如何用\nDeep Seek Harness设置它\n我认为这本身就是一个\n非常有趣的创新\n尤其是来自Deep Seek\n这是一个非常强大的框架\n我玩得很开心\n他们的核心思想是\n一切皆可插绑\n我为此做了一个视频\n我强烈推荐观看那个视频来入门\n好的\n但在此之前\n我会谈谈在哪里可以获取这个模型\n它已经在Hugging Face上可用了\n如果你有Apple Silicon,我推荐使用MLX版本。\n如果你想在Linux或Windows上运行,\nVLM和Stealing是两个可以尝试的选择。\n所以现在我正在DGX Park上,\n我将运行MVIP4版本。\n我有一个两节点的集群。\n现在DGX Park并不是真正为运行270亿密集模型设计的。\n所以我在单线程上\n大约每秒15到20个零排\n在并发情况下\n我想我能达到\n每秒60到70个零排\n这相当可用\n现在要运行这个模型\n你只需运行这个\n它会为你设置服务器\n你可以通过网络访问它\n这基本上会为你下载模型\n以及草稿模型\n它启用了推测解码\n好的,对推理,我已经下载了DeepSeek Harness。\n我为此做了一个视频,我强烈推荐观看它来了解基础知识。\n好的,这就是Harness的样子。\n现在为了设置Queen模型,我不得不进入设置。\n然后在设置中,你实际上可以配置任何你选择的模型。\n这会为你打开这个YAML文件\n现在在这个YAML文件中\n你需要定义你的提供者\n所以我一直在尝试不同的模型\n所以你可以看到这里\n有一个运行在那个集群上的\nDeepSeek vFlash ZR3.827BMLX版本\n那现在我正在使用这个配置\n它在两个DGX\nSpark上运行\n原因是\n我想要使用完整的上下文窗口\n并处理多个\n并发请求\n你可能已经看到了编码能力\n所以我要从我发现最有趣的\n一个开始\n所以我提供了这张图片\n这是一张参考图片\n这是一个多摩台模型\n这意味着它可以理解图像和文本\n它盛程的描述相当准确\n我要求它使用OpenCV\n这是一个用于计算机视觉的Python包\n让它决定做一些分析\n所以它实际上编写了自己的框架\n来分析幻雄图像\n由于这是相当开放的\n所以它做了几件事\n例如这里它在看边缘检测\n这是在做一些类型的脚点叠加\n它用来识别图像中的脚点\n现在这完全在DGX集群上本地运行\n感谢VDia\n赞助计算资源\n这些视觉语言模型\n非常擅长识别物体\n但它们在技术方面非常糟糕\n但是这个模型实际上让我惊讶\n所以我说你在这张图片中\n看到多少辆车\n他数出了图像中总共有26辆车\n所以我想他可能只是编了一个数字\n那然后他说好的\n从顶行开始\n基本上就是这里\n如果你能数一下车\n1 2 3 4 5 6 7 8 9 10 11\n对吧\n所以他数出了总共11辆车\n然后他谈到主停车场\n在那里\n你实际上可以看到\n有1,2,3,4,5,6,7,8,9辆车\n出于某种原因\n它没有把这排算作主停车场的一部分\n但技术结果相当准确\n为此我们需要看四位列\n看起来它能够将图像分成不同大小的补丁\n然后识别物体并在这些补丁内计数\n所以它不仅仅是识别物体\n还能指向它们\n我们在Deep Seek Vision模型论文中\n见过这种能力\n但还没有在现实中真正见过\n好的\n但然后它实际上变得更加有趣\n所以我说\n图片中有多少辆白色的车\n然后它再次开始像之前那样精确的技术\n所以它划分并看顶行\n然后主停车场\n然后路边\n根据它的说法\n总共有八辆不同的白色汽车\n如果你看原始图像\n似乎有一二三四五辆白色汽车\n但它肯定加上了这个\n这个\n我想还有这个\n但然后我说\n好的\n如果我要求它在物体周围画边界框呢\n但有趣的是\n看到实际结果\n为了做到这一点\n它实际上写了一个python程序\n所以你看到的边界框坐标\n这些直接来自模型\n所以模型实际上指向图像中的不同物体\n然后在上面绘制边界框\n好的\n所以这是结果\n实际上相当令人印象深刻\n尤其是这是一个视觉语言模型\n仅仅通过看图像就绘制边界框\n它们并不完美\n对吧\n所以如果你看这些\n可能覆盖了更大的区域\n但总的来说它非常准确\n有人可能会争论这两个\n甚至皮卡车的情况\n然而我认为你可以说他们通过了\n你可以添加另一个模型\n来做物体检测\n然后你可以用这个模型来判断\n这些检测到的物体是否准确\n但总的来说\n我真的印象深刻\n我还没有见过一个视觉语言模型\n在\n好的,所以我运行的其他一些测试\n这是实时国际空间站追踪器的结果\n它实际上是我见过的\n开源模型中最好的输出之一\n这里的工作真的令人印象深刻\n而且在实时更新方面相当准确\n它每5秒做一次\n完全按照我们的要求\n但输出的质量实际上将取决于你设置的推理努力\n接下来我们需要谈谈推理努力\n有四个不同的级别完全关闭\n这变成非推理模型\n然后低中超高没有高\n有时当你设置为超高或中等时\n它会将整个完成预算\n用作思考预算\n你不会得到任何输出\n我已经看到这种情况发生在我身上好几次了\n所以对于完全关闭\n我要求它使用网络搜索\n创建一个关于它自己的网站\n所以输出令牌大约是两万个令牌\n所以这个是完全关闭\n你实际上可以看到\n它只是编造了一些数字\n我要求模型进行网络搜索\n在一个他以前没见过的框架中\n然后创建一个网站解释一切\n然后仅仅通过将推理努力改为第一\n他实际上开始生成一些非常有趣的结果\n这实际上是一个相当不错的网站\n可以说他甚至比像GPT5.5最初发布时更好\n仍然有那种蓝色调\n但整体视觉效果非常好\n现在这是当你把游戏预算提高到中等时的结果\n它开始添加这些漂亮的动画\n我认为与完全关闭推理相比\n质量显著提高\n对,所以你可以看到显著的改进\n然而它使用的令牌数量\n与非思考版本相比大幅增加\n最后相同的提示\n相同的上下文是超高推理\n视觉质量相当不错\n动画与中等推理非常相似\n但视觉语言有点不同\n颜色与低和中等不同\n似乎推理与输出主题的外观有很大关系\n我认为这是一个相当可用的网站\n你可以从本地模型得到\n现在快速看一下令牌数量\n推理关闭\n输入令牌是8万2千个令牌\n输出大约2万个令牌\n花了大约15分钟\n我猜第二个大约是25\n但每次运行之间会变化\n因为它在运行多个并方模型\n对于超高\n几乎50万个输入令牌\n6万个输出令牌\n大约86分钟\n现在中等出于某种原因\n实际上消耗了更多的令牌\n所以几乎是8\n推理35分钟\n7万个输入令牌\n大约31000个输出令牌\n对\n所以你可以看到\n如果你提高推理努力\n令牌数量基本上会爆炸\n现在这发生在一个全新的框架下\n我认为代理以前没见过\n这是Leap Seek Harness\n但它有这个非常整洁的轨迹功能\n所以你可以点击这个\n这基本上会显示上下文中\n到目前为止发生的所有事情\n所以我们有初始系统提示\n一切都可以省计\n你实际上可以看到\n系统提示中具体包含了什么\n然后用户消息是什么\n这是我给系统的初始消息\n然后上下文\n这是模型对Deep Seek Harness的初始上下文\n其中一些与QDAV模型无关\n但它仍然会看到这些\n所以这些是所有可用的技能列表\n对吧\n但然后美妙之处在于\n你实际上可以看到模型\n采取了什么具体行动\n如果有工具调用\n你可以看到具体负载是什么\n模式是什么\n它使用了结果时间\n这相当不可思议\n就像我还没有见过\n任何其他框架提供这种级别的细节\n通常你需要使用一些\n其他追踪工具进行可观测性\n但Deepsea\n使你能够实际查看\n所有采取的行动以及时间线\n所以你可以\n省计它\n你可以找出代理在哪里失败\n然后如果你看到多个工具\n调用失败\n你甚至可以更新你的工具\n这相当不可思议\n现在我确实计划创建一个专门探索Deep-Sea Carnest\n中这个标记机制的视频\n我觉得这是对框架设计的一个非常新的看法\n很多人可以从中学习\n好的顺便说一下\n它刚刚超过了16万个Github新标\n不到一周大\n这太疯狂了\n目前我认为这是GitHub历史上增长最快的开源项目\n有趣的是他们不接受任何外部贡献\n所以这一切都要感谢DeepSeek团队\n好的,我将运行更多实验\n特别是视觉分析能力\n因为我认为我还没有见过这样的东西\n但一定要去看看\n如果你有任何具体问题,请告诉我\n无论如何\n我希望你觉得这个视频有用\n感谢观看\n像往常一样\n我们下个视频见", "language": "zh", "duration": 893.085938, "segments": [ { "start": 0.0, "end": 28.64, "text": "這是你可以在消費級硬件上運行的最好的本地模型。而且就其规模而言,它真的非常出色。在本視頻中,我將向你展示如何將其與DeepSeekHarness這樣強大的編碼代理結合,來構建強大的代理系統。而且它還有一些非常有趣的隱藏能力。例如,它在圖像理解和繪製邊界框方面非常出色。所以在本視頻中,我的目標是向你展示如何設置它。", "chunk": 0, "language": "zh", "avg_logprob": -0.14369327921262928, "no_speech_prob": 9.297830734500412e-12, "compression_ratio": 1.3510971786833856, "source": "whisper" }, { "start": 28.64, "end": 30.28, "text": "如果你想使用这个模型", "chunk": 0, "language": "zh", "avg_logprob": -0.12138793304676318, "no_speech_prob": 1.0203277459042148e-11, "compression_ratio": 1.2419354838709677, "source": "whisper" }, { "start": 30.28, "end": 32.18, "text": "应该使用哪些不同的配置", "chunk": 0, "language": "zh", "avg_logprob": -0.12138793304676318, "no_speech_prob": 1.0203277459042148e-11, "compression_ratio": 1.2419354838709677, "source": "whisper" }, { "start": 32.18, "end": 37.62, "text": "以及许多人在推理时常犯的错误", "chunk": 0, "language": "zh", "avg_logprob": -0.12138793304676318, "no_speech_prob": 1.0203277459042148e-11, "compression_ratio": 1.2419354838709677, "source": "whisper" }, { "start": 37.62, "end": 41.44, "text": "首先我们来看几个不同的基准测试", "chunk": 0, "language": "zh", "avg_logprob": -0.12138793304676318, "no_speech_prob": 1.0203277459042148e-11, "compression_ratio": 1.2419354838709677, "source": "whisper" }, { "start": 41.44, "end": 44.74, "text": "因为这确实能展示这个模型的强大之处", "chunk": 0, "language": "zh", "avg_logprob": -0.12138793304676318, "no_speech_prob": 1.0203277459042148e-11, "compression_ratio": 1.2419354838709677, "source": "whisper" }, { "start": 44.74, "end": 47.3, "text": "这是人工智能代理指数", "chunk": 0, "language": "zh", "avg_logprob": -0.12138793304676318, "no_speech_prob": 1.0203277459042148e-11, "compression_ratio": 1.2419354838709677, "source": "whisper" }, { "start": 47.3, "end": 50.34, "text": "目前是51", "chunk": 0, "language": "zh", "avg_logprob": -0.12138793304676318, "no_speech_prob": 1.0203277459042148e-11, "compression_ratio": 1.2419354838709677, "source": "whisper" }, { "start": 50.34, "end": 53.34, "text": "实际上仅落后于KIMIK3", "chunk": 0, "language": "zh", "avg_logprob": -0.12138793304676318, "no_speech_prob": 1.0203277459042148e-11, "compression_ratio": 1.2419354838709677, "source": "whisper" }, { "start": 53.34, "end": 57.62, "text": "KIMIK3是一个2.8万亿参数的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.12138793304676318, "no_speech_prob": 1.0203277459042148e-11, "compression_ratio": 1.2419354838709677, "source": "whisper" }, { "start": 57.62, "end": 61.84, "text": "最重要的是性能在真实世界测试中依然保持", "chunk": 0, "language": "zh", "avg_logprob": -0.09144271422769422, "no_speech_prob": 8.829503968244001e-12, "compression_ratio": 1.155844155844156, "source": "whisper" }, { "start": 61.84, "end": 64.72, "text": "稍后我会在视频中展示几个例子", "chunk": 0, "language": "zh", "avg_logprob": -0.09144271422769422, "no_speech_prob": 8.829503968244001e-12, "compression_ratio": 1.155844155844156, "source": "whisper" }, { "start": 64.72, "end": 66.84, "text": "现在来对比一下", "chunk": 0, "language": "zh", "avg_logprob": -0.09144271422769422, "no_speech_prob": 8.829503968244001e-12, "compression_ratio": 1.155844155844156, "source": "whisper" }, { "start": 66.84, "end": 72.6, "text": "它的智能和性能非常接近最高设置下的Cloud 4.8", "chunk": 0, "language": "zh", "avg_logprob": -0.09144271422769422, "no_speech_prob": 8.829503968244001e-12, "compression_ratio": 1.155844155844156, "source": "whisper" }, { "start": 72.6, "end": 79.62, "text": "而那是几个月前的顶尖模型", "chunk": 0, "language": "zh", "avg_logprob": -0.09144271422769422, "no_speech_prob": 8.829503968244001e-12, "compression_ratio": 1.155844155844156, "source": "whisper" }, { "start": 79.62, "end": 83.58, "text": "有人可能会说这可能是基准测试的结果", "chunk": 0, "language": "zh", "avg_logprob": -0.09144271422769422, "no_speech_prob": 8.829503968244001e-12, "compression_ratio": 1.155844155844156, "source": "whisper" }, { "start": 83.58, "end": 85.42, "text": "但真实世界的性能", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 85.42, "end": 87.96, "text": "实际上非常非常出色", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 87.96, "end": 89.5, "text": "我将向你展示如何用", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 89.5, "end": 91.3, "text": "Deep Seek Harness设置它", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 91.3, "end": 93.5, "text": "我认为这本身就是一个", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 93.5, "end": 94.54, "text": "非常有趣的创新", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 94.54, "end": 96.04, "text": "尤其是来自Deep Seek", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 96.04, "end": 97.94, "text": "这是一个非常强大的框架", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 97.94, "end": 99.12, "text": "我玩得很开心", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 99.12, "end": 102.12, "text": "他们的核心思想是", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 102.12, "end": 103.56, "text": "一切皆可插绑", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 103.56, "end": 105.16, "text": "我为此做了一个视频", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 105.16, "end": 107.64, "text": "我强烈推荐观看那个视频来入门", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 107.64, "end": 108.12, "text": "好的", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 108.12, "end": 109.06, "text": "但在此之前", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 109.06, "end": 111.2, "text": "我会谈谈在哪里可以获取这个模型", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 111.2, "end": 113.16, "text": "它已经在Hugging Face上可用了", "chunk": 0, "language": "zh", "avg_logprob": -0.10897925625676694, "no_speech_prob": 1.3648156389767241e-11, "compression_ratio": 1.2693409742120343, "source": "whisper" }, { "start": 113.16, "end": 115.98, "text": "如果你有Apple Silicon,我推荐使用MLX版本。", "chunk": 0, "language": "zh", "avg_logprob": -0.17235200107097626, "no_speech_prob": 1.3138170239235247e-11, "compression_ratio": 1.1320754716981132, "source": "whisper" }, { "start": 117.4, "end": 120.68, "text": "如果你想在Linux或Windows上运行,", "chunk": 0, "language": "zh", "avg_logprob": -0.17235200107097626, "no_speech_prob": 1.3138170239235247e-11, "compression_ratio": 1.1320754716981132, "source": "whisper" }, { "start": 120.88, "end": 124.5, "text": "VLM和Stealing是两个可以尝试的选择。", "chunk": 0, "language": "zh", "avg_logprob": -0.17235200107097626, "no_speech_prob": 1.3138170239235247e-11, "compression_ratio": 1.1320754716981132, "source": "whisper" }, { "start": 125.28, "end": 128.42, "text": "所以现在我正在DGX Park上,", "chunk": 0, "language": "zh", "avg_logprob": -0.17235200107097626, "no_speech_prob": 1.3138170239235247e-11, "compression_ratio": 1.1320754716981132, "source": "whisper" }, { "start": 128.52, "end": 130.76, "text": "我将运行MVIP4版本。", "chunk": 0, "language": "zh", "avg_logprob": -0.17235200107097626, "no_speech_prob": 1.3138170239235247e-11, "compression_ratio": 1.1320754716981132, "source": "whisper" }, { "start": 130.82, "end": 132.84, "text": "我有一个两节点的集群。", "chunk": 0, "language": "zh", "avg_logprob": -0.17235200107097626, "no_speech_prob": 1.3138170239235247e-11, "compression_ratio": 1.1320754716981132, "source": "whisper" }, { "start": 133.3, "end": 138.66, "text": "现在DGX Park并不是真正为运行270亿密集模型设计的。", "chunk": 0, "language": "zh", "avg_logprob": -0.17235200107097626, "no_speech_prob": 1.3138170239235247e-11, "compression_ratio": 1.1320754716981132, "source": "whisper" }, { "start": 138.66, "end": 141.34, "text": "所以我在单线程上", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 141.34, "end": 143.48, "text": "大约每秒15到20个零排", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 143.48, "end": 145.6, "text": "在并发情况下", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 145.6, "end": 146.9, "text": "我想我能达到", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 146.9, "end": 149.5, "text": "每秒60到70个零排", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 149.5, "end": 150.5, "text": "这相当可用", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 150.5, "end": 152.98, "text": "现在要运行这个模型", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 152.98, "end": 154.86, "text": "你只需运行这个", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 154.86, "end": 158.0, "text": "它会为你设置服务器", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 158.0, "end": 160.02, "text": "你可以通过网络访问它", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 160.02, "end": 163.1, "text": "这基本上会为你下载模型", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 163.1, "end": 164.4, "text": "以及草稿模型", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 164.4, "end": 167.7, "text": "它启用了推测解码", "chunk": 0, "language": "zh", "avg_logprob": -0.12066284015024309, "no_speech_prob": 1.0326546041883322e-11, "compression_ratio": 1.297872340425532, "source": "whisper" }, { "start": 167.7, "end": 170.58, "text": "好的,对推理,我已经下载了DeepSeek Harness。", "chunk": 0, "language": "zh", "avg_logprob": -0.1431444914444633, "no_speech_prob": 1.1246401379616522e-11, "compression_ratio": 1.1633466135458168, "source": "whisper" }, { "start": 171.54, "end": 176.58, "text": "我为此做了一个视频,我强烈推荐观看它来了解基础知识。", "chunk": 0, "language": "zh", "avg_logprob": -0.1431444914444633, "no_speech_prob": 1.1246401379616522e-11, "compression_ratio": 1.1633466135458168, "source": "whisper" }, { "start": 177.76, "end": 180.02, "text": "好的,这就是Harness的样子。", "chunk": 0, "language": "zh", "avg_logprob": -0.1431444914444633, "no_speech_prob": 1.1246401379616522e-11, "compression_ratio": 1.1633466135458168, "source": "whisper" }, { "start": 180.84, "end": 184.12, "text": "现在为了设置Queen模型,我不得不进入设置。", "chunk": 0, "language": "zh", "avg_logprob": -0.1431444914444633, "no_speech_prob": 1.1246401379616522e-11, "compression_ratio": 1.1633466135458168, "source": "whisper" }, { "start": 186.06, "end": 190.62, "text": "然后在设置中,你实际上可以配置任何你选择的模型。", "chunk": 0, "language": "zh", "avg_logprob": -0.1431444914444633, "no_speech_prob": 1.1246401379616522e-11, "compression_ratio": 1.1633466135458168, "source": "whisper" }, { "start": 190.62, "end": 197.98, "text": "这会为你打开这个YAML文件", "chunk": 0, "language": "zh", "avg_logprob": -0.1520971479870024, "no_speech_prob": 7.1009630467355755e-12, "compression_ratio": 1.1144067796610169, "source": "whisper" }, { "start": 197.98, "end": 201.02, "text": "现在在这个YAML文件中", "chunk": 0, "language": "zh", "avg_logprob": -0.1520971479870024, "no_speech_prob": 7.1009630467355755e-12, "compression_ratio": 1.1144067796610169, "source": "whisper" }, { "start": 201.02, "end": 202.74, "text": "你需要定义你的提供者", "chunk": 0, "language": "zh", "avg_logprob": -0.1520971479870024, "no_speech_prob": 7.1009630467355755e-12, "compression_ratio": 1.1144067796610169, "source": "whisper" }, { "start": 202.74, "end": 205.94, "text": "所以我一直在尝试不同的模型", "chunk": 0, "language": "zh", "avg_logprob": -0.1520971479870024, "no_speech_prob": 7.1009630467355755e-12, "compression_ratio": 1.1144067796610169, "source": "whisper" }, { "start": 205.94, "end": 207.04, "text": "所以你可以看到这里", "chunk": 0, "language": "zh", "avg_logprob": -0.1520971479870024, "no_speech_prob": 7.1009630467355755e-12, "compression_ratio": 1.1144067796610169, "source": "whisper" }, { "start": 207.04, "end": 209.02, "text": "有一个运行在那个集群上的", "chunk": 0, "language": "zh", "avg_logprob": -0.1520971479870024, "no_speech_prob": 7.1009630467355755e-12, "compression_ratio": 1.1144067796610169, "source": "whisper" }, { "start": 209.02, "end": 213.84, "text": "DeepSeek vFlash ZR3.827BMLX版本", "chunk": 0, "language": "zh", "avg_logprob": -0.1520971479870024, "no_speech_prob": 7.1009630467355755e-12, "compression_ratio": 1.1144067796610169, "source": "whisper" }, { "start": 213.84, "end": 217.7, "text": "那现在我正在使用这个配置", "chunk": 0, "language": "zh", "avg_logprob": -0.1520971479870024, "no_speech_prob": 7.1009630467355755e-12, "compression_ratio": 1.1144067796610169, "source": "whisper" }, { "start": 217.7, "end": 220.38, "text": "它在两个DGX", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 220.38, "end": 221.86, "text": "Spark上运行", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 221.86, "end": 224.12, "text": "原因是", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 224.12, "end": 226.86, "text": "我想要使用完整的上下文窗口", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 226.86, "end": 228.44, "text": "并处理多个", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 228.44, "end": 229.5, "text": "并发请求", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 229.5, "end": 231.92, "text": "你可能已经看到了编码能力", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 231.92, "end": 234.82, "text": "所以我要从我发现最有趣的", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 234.82, "end": 235.46, "text": "一个开始", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 235.46, "end": 237.14, "text": "所以我提供了这张图片", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 237.14, "end": 238.52, "text": "这是一张参考图片", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 238.52, "end": 240.08, "text": "这是一个多摩台模型", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 240.08, "end": 243.06, "text": "这意味着它可以理解图像和文本", "chunk": 0, "language": "zh", "avg_logprob": -0.14687870488022314, "no_speech_prob": 9.050970910251532e-12, "compression_ratio": 1.2292490118577075, "source": "whisper" }, { "start": 247.7, "end": 255.1, "text": "它盛程的描述相当准确", "chunk": 0, "language": "zh", "avg_logprob": -0.11976419067382812, "no_speech_prob": 8.791013923759028e-12, "compression_ratio": 1.1567796610169492, "source": "whisper" }, { "start": 255.1, "end": 258.66, "text": "我要求它使用OpenCV", "chunk": 0, "language": "zh", "avg_logprob": -0.11976419067382812, "no_speech_prob": 8.791013923759028e-12, "compression_ratio": 1.1567796610169492, "source": "whisper" }, { "start": 258.66, "end": 261.84, "text": "这是一个用于计算机视觉的Python包", "chunk": 0, "language": "zh", "avg_logprob": -0.11976419067382812, "no_speech_prob": 8.791013923759028e-12, "compression_ratio": 1.1567796610169492, "source": "whisper" }, { "start": 261.84, "end": 263.96, "text": "让它决定做一些分析", "chunk": 0, "language": "zh", "avg_logprob": -0.11976419067382812, "no_speech_prob": 8.791013923759028e-12, "compression_ratio": 1.1567796610169492, "source": "whisper" }, { "start": 263.96, "end": 267.24, "text": "所以它实际上编写了自己的框架", "chunk": 0, "language": "zh", "avg_logprob": -0.11976419067382812, "no_speech_prob": 8.791013923759028e-12, "compression_ratio": 1.1567796610169492, "source": "whisper" }, { "start": 267.24, "end": 269.22, "text": "来分析幻雄图像", "chunk": 0, "language": "zh", "avg_logprob": -0.11976419067382812, "no_speech_prob": 8.791013923759028e-12, "compression_ratio": 1.1567796610169492, "source": "whisper" }, { "start": 269.22, "end": 271.7, "text": "由于这是相当开放的", "chunk": 0, "language": "zh", "avg_logprob": -0.11976419067382812, "no_speech_prob": 8.791013923759028e-12, "compression_ratio": 1.1567796610169492, "source": "whisper" }, { "start": 271.7, "end": 272.96, "text": "所以它做了几件事", "chunk": 0, "language": "zh", "avg_logprob": -0.11976419067382812, "no_speech_prob": 8.791013923759028e-12, "compression_ratio": 1.1567796610169492, "source": "whisper" }, { "start": 272.96, "end": 276.34, "text": "例如这里它在看边缘检测", "chunk": 0, "language": "zh", "avg_logprob": -0.11976419067382812, "no_speech_prob": 8.791013923759028e-12, "compression_ratio": 1.1567796610169492, "source": "whisper" }, { "start": 276.34, "end": 280.04, "text": "这是在做一些类型的脚点叠加", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 280.04, "end": 282.6, "text": "它用来识别图像中的脚点", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 282.6, "end": 286.06, "text": "现在这完全在DGX集群上本地运行", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 286.06, "end": 287.18, "text": "感谢VDia", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 287.18, "end": 288.84, "text": "赞助计算资源", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 288.84, "end": 292.28, "text": "这些视觉语言模型", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 292.28, "end": 294.06, "text": "非常擅长识别物体", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 294.06, "end": 296.54, "text": "但它们在技术方面非常糟糕", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 296.54, "end": 300.54, "text": "但是这个模型实际上让我惊讶", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 300.54, "end": 302.54, "text": "所以我说你在这张图片中", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 302.54, "end": 303.8, "text": "看到多少辆车", "chunk": 0, "language": "zh", "avg_logprob": -0.12935443691440396, "no_speech_prob": 1.1881683137371368e-11, "compression_ratio": 1.2153846153846153, "source": "whisper" }, { "start": 303.8, "end": 306.62, "text": "他数出了图像中总共有26辆车", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 306.62, "end": 311.82, "text": "所以我想他可能只是编了一个数字", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 311.82, "end": 313.66, "text": "那然后他说好的", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 313.66, "end": 315.0, "text": "从顶行开始", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 315.0, "end": 316.82, "text": "基本上就是这里", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 316.82, "end": 318.54, "text": "如果你能数一下车", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 318.54, "end": 323.38, "text": "1 2 3 4 5 6 7 8 9 10 11", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 323.38, "end": 323.98, "text": "对吧", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 323.98, "end": 326.24, "text": "所以他数出了总共11辆车", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 326.24, "end": 328.74, "text": "然后他谈到主停车场", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 328.74, "end": 329.36, "text": "在那里", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 329.36, "end": 330.7, "text": "你实际上可以看到", "chunk": 0, "language": "zh", "avg_logprob": -0.11710954116562665, "no_speech_prob": 1.0772441098871877e-11, "compression_ratio": 1.2337662337662338, "source": "whisper" }, { "start": 330.7, "end": 335.48, "text": "有1,2,3,4,5,6,7,8,9辆车", "chunk": 0, "language": "zh", "avg_logprob": -0.08555197311659991, "no_speech_prob": 7.184255360753733e-12, "compression_ratio": 1.117117117117117, "source": "whisper" }, { "start": 335.48, "end": 336.8, "text": "出于某种原因", "chunk": 0, "language": "zh", "avg_logprob": -0.08555197311659991, "no_speech_prob": 7.184255360753733e-12, "compression_ratio": 1.117117117117117, "source": "whisper" }, { "start": 336.8, "end": 340.02, "text": "它没有把这排算作主停车场的一部分", "chunk": 0, "language": "zh", "avg_logprob": -0.08555197311659991, "no_speech_prob": 7.184255360753733e-12, "compression_ratio": 1.117117117117117, "source": "whisper" }, { "start": 340.02, "end": 342.02, "text": "但技术结果相当准确", "chunk": 0, "language": "zh", "avg_logprob": -0.08555197311659991, "no_speech_prob": 7.184255360753733e-12, "compression_ratio": 1.117117117117117, "source": "whisper" }, { "start": 342.02, "end": 346.96, "text": "为此我们需要看四位列", "chunk": 0, "language": "zh", "avg_logprob": -0.08555197311659991, "no_speech_prob": 7.184255360753733e-12, "compression_ratio": 1.117117117117117, "source": "whisper" }, { "start": 346.96, "end": 350.68, "text": "看起来它能够将图像分成不同大小的补丁", "chunk": 0, "language": "zh", "avg_logprob": -0.08555197311659991, "no_speech_prob": 7.184255360753733e-12, "compression_ratio": 1.117117117117117, "source": "whisper" }, { "start": 350.68, "end": 354.48, "text": "然后识别物体并在这些补丁内计数", "chunk": 0, "language": "zh", "avg_logprob": -0.08555197311659991, "no_speech_prob": 7.184255360753733e-12, "compression_ratio": 1.117117117117117, "source": "whisper" }, { "start": 360.7, "end": 362.66, "text": "所以它不仅仅是识别物体", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 362.66, "end": 364.12, "text": "还能指向它们", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 364.12, "end": 367.2, "text": "我们在Deep Seek Vision模型论文中", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 367.2, "end": 368.64, "text": "见过这种能力", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 368.64, "end": 372.84, "text": "但还没有在现实中真正见过", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 372.84, "end": 373.08, "text": "好的", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 373.08, "end": 374.96, "text": "但然后它实际上变得更加有趣", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 374.96, "end": 375.86, "text": "所以我说", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 375.86, "end": 378.3, "text": "图片中有多少辆白色的车", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 378.3, "end": 381.68, "text": "然后它再次开始像之前那样精确的技术", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 381.68, "end": 384.26, "text": "所以它划分并看顶行", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 384.26, "end": 385.72, "text": "然后主停车场", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 385.72, "end": 386.7, "text": "然后路边", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 386.7, "end": 389.5, "text": "根据它的说法", "chunk": 0, "language": "zh", "avg_logprob": -0.1147674886803878, "no_speech_prob": 1.3895784696515978e-11, "compression_ratio": 1.227891156462585, "source": "whisper" }, { "start": 389.5, "end": 392.46, "text": "总共有八辆不同的白色汽车", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 392.46, "end": 395.1, "text": "如果你看原始图像", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 395.1, "end": 398.88, "text": "似乎有一二三四五辆白色汽车", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 398.88, "end": 401.74, "text": "但它肯定加上了这个", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 401.74, "end": 402.32, "text": "这个", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 402.32, "end": 403.86, "text": "我想还有这个", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 403.86, "end": 405.94, "text": "但然后我说", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 405.94, "end": 406.36, "text": "好的", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 406.36, "end": 409.6, "text": "如果我要求它在物体周围画边界框呢", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 409.6, "end": 411.86, "text": "但有趣的是", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 411.86, "end": 413.46, "text": "看到实际结果", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 413.46, "end": 415.1, "text": "为了做到这一点", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 415.1, "end": 416.92, "text": "它实际上写了一个python程序", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 416.92, "end": 418.5, "text": "所以你看到的边界框坐标", "chunk": 0, "language": "zh", "avg_logprob": -0.1258601744969686, "no_speech_prob": 9.749425225469288e-12, "compression_ratio": 1.3003802281368821, "source": "whisper" }, { "start": 418.5, "end": 420.24, "text": "这些直接来自模型", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 420.24, "end": 425.42, "text": "所以模型实际上指向图像中的不同物体", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 425.42, "end": 427.38, "text": "然后在上面绘制边界框", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 427.38, "end": 431.26, "text": "好的", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 431.26, "end": 432.76, "text": "所以这是结果", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 432.76, "end": 434.46, "text": "实际上相当令人印象深刻", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 434.46, "end": 437.86, "text": "尤其是这是一个视觉语言模型", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 437.86, "end": 440.66, "text": "仅仅通过看图像就绘制边界框", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 440.66, "end": 441.58, "text": "它们并不完美", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 441.58, "end": 442.06, "text": "对吧", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 442.06, "end": 443.68, "text": "所以如果你看这些", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 443.68, "end": 445.54, "text": "可能覆盖了更大的区域", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 445.54, "end": 448.08, "text": "但总的来说它非常准确", "chunk": 0, "language": "zh", "avg_logprob": -0.10751003761814065, "no_speech_prob": 1.0275143583204915e-11, "compression_ratio": 1.3384615384615384, "source": "whisper" }, { "start": 448.5, "end": 450.26, "text": "有人可能会争论这两个", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 450.26, "end": 451.8, "text": "甚至皮卡车的情况", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 451.8, "end": 455.46, "text": "然而我认为你可以说他们通过了", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 455.46, "end": 458.02, "text": "你可以添加另一个模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 458.02, "end": 459.26, "text": "来做物体检测", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 459.26, "end": 462.7, "text": "然后你可以用这个模型来判断", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 462.7, "end": 464.56, "text": "这些检测到的物体是否准确", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 464.56, "end": 468.62, "text": "但总的来说", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 468.62, "end": 470.4, "text": "我真的印象深刻", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 470.4, "end": 473.08, "text": "我还没有见过一个视觉语言模型", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 473.08, "end": 473.58, "text": "在", "chunk": 0, "language": "zh", "avg_logprob": -0.11176216503805365, "no_speech_prob": 1.0441341367906087e-11, "compression_ratio": 1.3274336283185841, "source": "whisper" }, { "start": 487.27, "end": 491.39, "text": "好的,所以我运行的其他一些测试", "chunk": 1, "language": "zh", "avg_logprob": -0.12077047559950087, "no_speech_prob": 9.03709225508198e-12, "compression_ratio": 1.2565055762081785, "source": "whisper" }, { "start": 491.39, "end": 494.43, "text": "这是实时国际空间站追踪器的结果", "chunk": 1, "language": "zh", "avg_logprob": -0.12077047559950087, "no_speech_prob": 9.03709225508198e-12, "compression_ratio": 1.2565055762081785, "source": "whisper" }, { "start": 494.43, "end": 496.67, "text": "它实际上是我见过的", "chunk": 1, "language": "zh", "avg_logprob": -0.12077047559950087, "no_speech_prob": 9.03709225508198e-12, "compression_ratio": 1.2565055762081785, "source": "whisper" }, { "start": 496.67, "end": 499.41, "text": "开源模型中最好的输出之一", "chunk": 1, "language": "zh", "avg_logprob": -0.12077047559950087, "no_speech_prob": 9.03709225508198e-12, "compression_ratio": 1.2565055762081785, "source": "whisper" }, { "start": 499.41, "end": 501.77, "text": "这里的工作真的令人印象深刻", "chunk": 1, "language": "zh", "avg_logprob": -0.12077047559950087, "no_speech_prob": 9.03709225508198e-12, "compression_ratio": 1.2565055762081785, "source": "whisper" }, { "start": 501.77, "end": 504.47, "text": "而且在实时更新方面相当准确", "chunk": 1, "language": "zh", "avg_logprob": -0.12077047559950087, "no_speech_prob": 9.03709225508198e-12, "compression_ratio": 1.2565055762081785, "source": "whisper" }, { "start": 504.47, "end": 507.41, "text": "它每5秒做一次", "chunk": 1, "language": "zh", "avg_logprob": -0.12077047559950087, "no_speech_prob": 9.03709225508198e-12, "compression_ratio": 1.2565055762081785, "source": "whisper" }, { "start": 507.41, "end": 508.93, "text": "完全按照我们的要求", "chunk": 1, "language": "zh", "avg_logprob": -0.12077047559950087, "no_speech_prob": 9.03709225508198e-12, "compression_ratio": 1.2565055762081785, "source": "whisper" }, { "start": 508.93, "end": 513.65, "text": "但输出的质量实际上将取决于你设置的推理努力", "chunk": 1, "language": "zh", "avg_logprob": -0.12077047559950087, "no_speech_prob": 9.03709225508198e-12, "compression_ratio": 1.2565055762081785, "source": "whisper" }, { "start": 513.65, "end": 519.13, "text": "接下来我们需要谈谈推理努力", "chunk": 1, "language": "zh", "avg_logprob": -0.14259791133379696, "no_speech_prob": 9.946526241533249e-12, "compression_ratio": 1.2030456852791878, "source": "whisper" }, { "start": 519.13, "end": 521.31, "text": "有四个不同的级别完全关闭", "chunk": 1, "language": "zh", "avg_logprob": -0.14259791133379696, "no_speech_prob": 9.946526241533249e-12, "compression_ratio": 1.2030456852791878, "source": "whisper" }, { "start": 521.31, "end": 522.99, "text": "这变成非推理模型", "chunk": 1, "language": "zh", "avg_logprob": -0.14259791133379696, "no_speech_prob": 9.946526241533249e-12, "compression_ratio": 1.2030456852791878, "source": "whisper" }, { "start": 522.99, "end": 526.75, "text": "然后低中超高没有高", "chunk": 1, "language": "zh", "avg_logprob": -0.14259791133379696, "no_speech_prob": 9.946526241533249e-12, "compression_ratio": 1.2030456852791878, "source": "whisper" }, { "start": 535.35, "end": 537.79, "text": "有时当你设置为超高或中等时", "chunk": 1, "language": "zh", "avg_logprob": -0.14259791133379696, "no_speech_prob": 9.946526241533249e-12, "compression_ratio": 1.2030456852791878, "source": "whisper" }, { "start": 537.79, "end": 539.71, "text": "它会将整个完成预算", "chunk": 1, "language": "zh", "avg_logprob": -0.14259791133379696, "no_speech_prob": 9.946526241533249e-12, "compression_ratio": 1.2030456852791878, "source": "whisper" }, { "start": 539.71, "end": 540.81, "text": "用作思考预算", "chunk": 1, "language": "zh", "avg_logprob": -0.14259791133379696, "no_speech_prob": 9.946526241533249e-12, "compression_ratio": 1.2030456852791878, "source": "whisper" }, { "start": 540.81, "end": 542.51, "text": "你不会得到任何输出", "chunk": 1, "language": "zh", "avg_logprob": -0.14259791133379696, "no_speech_prob": 9.946526241533249e-12, "compression_ratio": 1.2030456852791878, "source": "whisper" }, { "start": 542.51, "end": 548.73, "text": "我已经看到这种情况发生在我身上好几次了", "chunk": 1, "language": "zh", "avg_logprob": -0.08719072109315454, "no_speech_prob": 7.615463815113799e-12, "compression_ratio": 1.3407079646017699, "source": "whisper" }, { "start": 548.73, "end": 550.23, "text": "所以对于完全关闭", "chunk": 1, "language": "zh", "avg_logprob": -0.08719072109315454, "no_speech_prob": 7.615463815113799e-12, "compression_ratio": 1.3407079646017699, "source": "whisper" }, { "start": 550.23, "end": 552.41, "text": "我要求它使用网络搜索", "chunk": 1, "language": "zh", "avg_logprob": -0.08719072109315454, "no_speech_prob": 7.615463815113799e-12, "compression_ratio": 1.3407079646017699, "source": "whisper" }, { "start": 552.41, "end": 555.51, "text": "创建一个关于它自己的网站", "chunk": 1, "language": "zh", "avg_logprob": -0.08719072109315454, "no_speech_prob": 7.615463815113799e-12, "compression_ratio": 1.3407079646017699, "source": "whisper" }, { "start": 555.51, "end": 558.55, "text": "所以输出令牌大约是两万个令牌", "chunk": 1, "language": "zh", "avg_logprob": -0.08719072109315454, "no_speech_prob": 7.615463815113799e-12, "compression_ratio": 1.3407079646017699, "source": "whisper" }, { "start": 558.55, "end": 565.21, "text": "所以这个是完全关闭", "chunk": 1, "language": "zh", "avg_logprob": -0.08719072109315454, "no_speech_prob": 7.615463815113799e-12, "compression_ratio": 1.3407079646017699, "source": "whisper" }, { "start": 565.21, "end": 567.21, "text": "你实际上可以看到", "chunk": 1, "language": "zh", "avg_logprob": -0.08719072109315454, "no_speech_prob": 7.615463815113799e-12, "compression_ratio": 1.3407079646017699, "source": "whisper" }, { "start": 567.21, "end": 569.45, "text": "它只是编造了一些数字", "chunk": 1, "language": "zh", "avg_logprob": -0.08719072109315454, "no_speech_prob": 7.615463815113799e-12, "compression_ratio": 1.3407079646017699, "source": "whisper" }, { "start": 569.45, "end": 571.77, "text": "我要求模型进行网络搜索", "chunk": 1, "language": "zh", "avg_logprob": -0.08719072109315454, "no_speech_prob": 7.615463815113799e-12, "compression_ratio": 1.3407079646017699, "source": "whisper" }, { "start": 571.77, "end": 574.27, "text": "在一个他以前没见过的框架中", "chunk": 1, "language": "zh", "avg_logprob": -0.12054813021705264, "no_speech_prob": 7.355898876126865e-12, "compression_ratio": 1.1733333333333333, "source": "whisper" }, { "start": 574.27, "end": 576.85, "text": "然后创建一个网站解释一切", "chunk": 1, "language": "zh", "avg_logprob": -0.12054813021705264, "no_speech_prob": 7.355898876126865e-12, "compression_ratio": 1.1733333333333333, "source": "whisper" }, { "start": 576.85, "end": 587.93, "text": "然后仅仅通过将推理努力改为第一", "chunk": 1, "language": "zh", "avg_logprob": -0.12054813021705264, "no_speech_prob": 7.355898876126865e-12, "compression_ratio": 1.1733333333333333, "source": "whisper" }, { "start": 587.93, "end": 592.13, "text": "他实际上开始生成一些非常有趣的结果", "chunk": 1, "language": "zh", "avg_logprob": -0.12054813021705264, "no_speech_prob": 7.355898876126865e-12, "compression_ratio": 1.1733333333333333, "source": "whisper" }, { "start": 592.13, "end": 594.55, "text": "这实际上是一个相当不错的网站", "chunk": 1, "language": "zh", "avg_logprob": -0.12054813021705264, "no_speech_prob": 7.355898876126865e-12, "compression_ratio": 1.1733333333333333, "source": "whisper" }, { "start": 594.55, "end": 598.79, "text": "可以说他甚至比像GPT5.5最初发布时更好", "chunk": 1, "language": "zh", "avg_logprob": -0.12054813021705264, "no_speech_prob": 7.355898876126865e-12, "compression_ratio": 1.1733333333333333, "source": "whisper" }, { "start": 601.77, "end": 604.01, "text": "仍然有那种蓝色调", "chunk": 1, "language": "zh", "avg_logprob": -0.1286163330078125, "no_speech_prob": 1.1152777486256316e-11, "compression_ratio": 1.220472440944882, "source": "whisper" }, { "start": 604.01, "end": 605.81, "text": "但整体视觉效果非常好", "chunk": 1, "language": "zh", "avg_logprob": -0.1286163330078125, "no_speech_prob": 1.1152777486256316e-11, "compression_ratio": 1.220472440944882, "source": "whisper" }, { "start": 605.81, "end": 612.91, "text": "现在这是当你把游戏预算提高到中等时的结果", "chunk": 1, "language": "zh", "avg_logprob": -0.1286163330078125, "no_speech_prob": 1.1152777486256316e-11, "compression_ratio": 1.220472440944882, "source": "whisper" }, { "start": 612.91, "end": 616.55, "text": "它开始添加这些漂亮的动画", "chunk": 1, "language": "zh", "avg_logprob": -0.1286163330078125, "no_speech_prob": 1.1152777486256316e-11, "compression_ratio": 1.220472440944882, "source": "whisper" }, { "start": 616.55, "end": 620.07, "text": "我认为与完全关闭推理相比", "chunk": 1, "language": "zh", "avg_logprob": -0.1286163330078125, "no_speech_prob": 1.1152777486256316e-11, "compression_ratio": 1.220472440944882, "source": "whisper" }, { "start": 620.07, "end": 621.69, "text": "质量显著提高", "chunk": 1, "language": "zh", "avg_logprob": -0.1286163330078125, "no_speech_prob": 1.1152777486256316e-11, "compression_ratio": 1.220472440944882, "source": "whisper" }, { "start": 621.69, "end": 625.55, "text": "对,所以你可以看到显著的改进", "chunk": 1, "language": "zh", "avg_logprob": -0.1286163330078125, "no_speech_prob": 1.1152777486256316e-11, "compression_ratio": 1.220472440944882, "source": "whisper" }, { "start": 625.55, "end": 627.97, "text": "然而它使用的令牌数量", "chunk": 1, "language": "zh", "avg_logprob": -0.1286163330078125, "no_speech_prob": 1.1152777486256316e-11, "compression_ratio": 1.220472440944882, "source": "whisper" }, { "start": 627.97, "end": 631.19, "text": "与非思考版本相比大幅增加", "chunk": 1, "language": "zh", "avg_logprob": -0.1286163330078125, "no_speech_prob": 1.1152777486256316e-11, "compression_ratio": 1.220472440944882, "source": "whisper" }, { "start": 631.77, "end": 635.07, "text": "最后相同的提示", "chunk": 1, "language": "zh", "avg_logprob": -0.1293615765041775, "no_speech_prob": 9.683545632022117e-12, "compression_ratio": 1.1956521739130435, "source": "whisper" }, { "start": 635.07, "end": 636.95, "text": "相同的上下文是超高推理", "chunk": 1, "language": "zh", "avg_logprob": -0.1293615765041775, "no_speech_prob": 9.683545632022117e-12, "compression_ratio": 1.1956521739130435, "source": "whisper" }, { "start": 642.95, "end": 644.97, "text": "视觉质量相当不错", "chunk": 1, "language": "zh", "avg_logprob": -0.1293615765041775, "no_speech_prob": 9.683545632022117e-12, "compression_ratio": 1.1956521739130435, "source": "whisper" }, { "start": 644.97, "end": 647.41, "text": "动画与中等推理非常相似", "chunk": 1, "language": "zh", "avg_logprob": -0.1293615765041775, "no_speech_prob": 9.683545632022117e-12, "compression_ratio": 1.1956521739130435, "source": "whisper" }, { "start": 647.41, "end": 649.35, "text": "但视觉语言有点不同", "chunk": 1, "language": "zh", "avg_logprob": -0.1293615765041775, "no_speech_prob": 9.683545632022117e-12, "compression_ratio": 1.1956521739130435, "source": "whisper" }, { "start": 649.35, "end": 656.83, "text": "颜色与低和中等不同", "chunk": 1, "language": "zh", "avg_logprob": -0.1293615765041775, "no_speech_prob": 9.683545632022117e-12, "compression_ratio": 1.1956521739130435, "source": "whisper" }, { "start": 656.83, "end": 670.17, "text": "似乎推理与输出主题的外观有很大关系", "chunk": 1, "language": "zh", "avg_logprob": -0.1193121402494369, "no_speech_prob": 6.09690076203151e-12, "compression_ratio": 1.054054054054054, "source": "whisper" }, { "start": 670.17, "end": 676.59, "text": "我认为这是一个相当可用的网站", "chunk": 1, "language": "zh", "avg_logprob": -0.1193121402494369, "no_speech_prob": 6.09690076203151e-12, "compression_ratio": 1.054054054054054, "source": "whisper" }, { "start": 676.59, "end": 679.11, "text": "你可以从本地模型得到", "chunk": 1, "language": "zh", "avg_logprob": -0.1193121402494369, "no_speech_prob": 6.09690076203151e-12, "compression_ratio": 1.054054054054054, "source": "whisper" }, { "start": 682.31, "end": 684.79, "text": "现在快速看一下令牌数量", "chunk": 1, "language": "zh", "avg_logprob": -0.1193121402494369, "no_speech_prob": 6.09690076203151e-12, "compression_ratio": 1.054054054054054, "source": "whisper" }, { "start": 684.79, "end": 687.47, "text": "推理关闭", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 687.47, "end": 690.09, "text": "输入令牌是8万2千个令牌", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 690.09, "end": 691.91, "text": "输出大约2万个令牌", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 691.91, "end": 693.13, "text": "花了大约15分钟", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 693.13, "end": 697.79, "text": "我猜第二个大约是25", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 697.79, "end": 699.61, "text": "但每次运行之间会变化", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 699.61, "end": 701.87, "text": "因为它在运行多个并方模型", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 701.87, "end": 705.83, "text": "对于超高", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 705.83, "end": 707.85, "text": "几乎50万个输入令牌", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 707.85, "end": 709.67, "text": "6万个输出令牌", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 709.67, "end": 711.21, "text": "大约86分钟", "chunk": 1, "language": "zh", "avg_logprob": -0.11101523179274339, "no_speech_prob": 8.459270610383651e-12, "compression_ratio": 1.3548387096774193, "source": "whisper" }, { "start": 714.79, "end": 717.21, "text": "现在中等出于某种原因", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 717.21, "end": 718.73, "text": "实际上消耗了更多的令牌", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 718.73, "end": 720.15, "text": "所以几乎是8", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 720.15, "end": 731.55, "text": "推理35分钟", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 731.55, "end": 732.85, "text": "7万个输入令牌", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 732.85, "end": 738.17, "text": "大约31000个输出令牌", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 738.17, "end": 738.83, "text": "对", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 738.83, "end": 739.65, "text": "所以你可以看到", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 739.65, "end": 740.95, "text": "如果你提高推理努力", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 740.95, "end": 743.07, "text": "令牌数量基本上会爆炸", "chunk": 1, "language": "zh", "avg_logprob": -0.1320293634244711, "no_speech_prob": 8.973981280302468e-12, "compression_ratio": 1.190217391304348, "source": "whisper" }, { "start": 743.07, "end": 747.07, "text": "现在这发生在一个全新的框架下", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 747.07, "end": 751.77, "text": "我认为代理以前没见过", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 751.77, "end": 752.99, "text": "这是Leap Seek Harness", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 752.99, "end": 755.99, "text": "但它有这个非常整洁的轨迹功能", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 755.99, "end": 757.27, "text": "所以你可以点击这个", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 757.27, "end": 759.91, "text": "这基本上会显示上下文中", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 759.91, "end": 763.57, "text": "到目前为止发生的所有事情", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 763.57, "end": 765.69, "text": "所以我们有初始系统提示", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 765.69, "end": 767.39, "text": "一切都可以省计", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 767.39, "end": 769.15, "text": "你实际上可以看到", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 769.15, "end": 771.23, "text": "系统提示中具体包含了什么", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 771.23, "end": 773.05, "text": "然后用户消息是什么", "chunk": 1, "language": "zh", "avg_logprob": -0.12070320712195502, "no_speech_prob": 1.1226040930178982e-11, "compression_ratio": 1.2677966101694915, "source": "whisper" }, { "start": 773.05, "end": 776.35, "text": "这是我给系统的初始消息", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 776.35, "end": 778.03, "text": "然后上下文", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 778.03, "end": 782.07, "text": "这是模型对Deep Seek Harness的初始上下文", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 782.07, "end": 785.31, "text": "其中一些与QDAV模型无关", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 785.31, "end": 787.07, "text": "但它仍然会看到这些", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 787.07, "end": 790.59, "text": "所以这些是所有可用的技能列表", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 790.59, "end": 791.11, "text": "对吧", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 791.11, "end": 792.89, "text": "但然后美妙之处在于", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 792.89, "end": 794.69, "text": "你实际上可以看到模型", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 794.69, "end": 796.47, "text": "采取了什么具体行动", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 796.47, "end": 798.49, "text": "如果有工具调用", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 798.49, "end": 800.47, "text": "你可以看到具体负载是什么", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 800.47, "end": 802.35, "text": "模式是什么", "chunk": 1, "language": "zh", "avg_logprob": -0.13288633973448427, "no_speech_prob": 9.837068659646064e-12, "compression_ratio": 1.313868613138686, "source": "whisper" }, { "start": 802.35, "end": 804.25, "text": "它使用了结果时间", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 804.25, "end": 805.87, "text": "这相当不可思议", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 805.87, "end": 807.87, "text": "就像我还没有见过", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 807.87, "end": 810.35, "text": "任何其他框架提供这种级别的细节", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 810.35, "end": 811.89, "text": "通常你需要使用一些", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 811.89, "end": 814.49, "text": "其他追踪工具进行可观测性", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 814.49, "end": 816.93, "text": "但Deepsea", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 816.93, "end": 818.47, "text": "使你能够实际查看", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 818.47, "end": 820.89, "text": "所有采取的行动以及时间线", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 820.89, "end": 824.11, "text": "所以你可以", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 824.11, "end": 825.01, "text": "省计它", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 825.01, "end": 827.21, "text": "你可以找出代理在哪里失败", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 827.21, "end": 829.79, "text": "然后如果你看到多个工具", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 829.79, "end": 830.85, "text": "调用失败", "chunk": 1, "language": "zh", "avg_logprob": -0.12198299407958985, "no_speech_prob": 1.0595179247840925e-11, "compression_ratio": 1.2661870503597121, "source": "whisper" }, { "start": 830.85, "end": 833.27, "text": "你甚至可以更新你的工具", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 833.27, "end": 835.07, "text": "这相当不可思议", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 835.07, "end": 839.39, "text": "现在我确实计划创建一个专门探索Deep-Sea Carnest", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 839.39, "end": 841.71, "text": "中这个标记机制的视频", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 841.71, "end": 845.39, "text": "我觉得这是对框架设计的一个非常新的看法", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 845.39, "end": 847.33, "text": "很多人可以从中学习", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 847.33, "end": 850.41, "text": "好的顺便说一下", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 850.41, "end": 853.75, "text": "它刚刚超过了16万个Github新标", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 853.75, "end": 855.65, "text": "不到一周大", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 855.65, "end": 857.05, "text": "这太疯狂了", "chunk": 1, "language": "zh", "avg_logprob": -0.16336720784505207, "no_speech_prob": 1.1551485462613087e-11, "compression_ratio": 1.1397849462365592, "source": "whisper" }, { "start": 857.05, "end": 864.13, "text": "目前我认为这是GitHub历史上增长最快的开源项目", "chunk": 1, "language": "zh", "avg_logprob": -0.133342200178441, "no_speech_prob": 9.464740623188472e-12, "compression_ratio": 1.1474820143884892, "source": "whisper" }, { "start": 864.13, "end": 869.31, "text": "有趣的是他们不接受任何外部贡献", "chunk": 1, "language": "zh", "avg_logprob": -0.133342200178441, "no_speech_prob": 9.464740623188472e-12, "compression_ratio": 1.1474820143884892, "source": "whisper" }, { "start": 869.31, "end": 874.03, "text": "所以这一切都要感谢DeepSeek团队", "chunk": 1, "language": "zh", "avg_logprob": -0.133342200178441, "no_speech_prob": 9.464740623188472e-12, "compression_ratio": 1.1474820143884892, "source": "whisper" }, { "start": 874.03, "end": 877.27, "text": "好的,我将运行更多实验", "chunk": 1, "language": "zh", "avg_logprob": -0.133342200178441, "no_speech_prob": 9.464740623188472e-12, "compression_ratio": 1.1474820143884892, "source": "whisper" }, { "start": 877.27, "end": 879.01, "text": "特别是视觉分析能力", "chunk": 1, "language": "zh", "avg_logprob": -0.133342200178441, "no_speech_prob": 9.464740623188472e-12, "compression_ratio": 1.1474820143884892, "source": "whisper" }, { "start": 879.01, "end": 880.95, "text": "因为我认为我还没有见过这样的东西", "chunk": 1, "language": "zh", "avg_logprob": -0.133342200178441, "no_speech_prob": 9.464740623188472e-12, "compression_ratio": 1.1474820143884892, "source": "whisper" }, { "start": 880.95, "end": 884.07, "text": "但一定要去看看", "chunk": 1, "language": "zh", "avg_logprob": -0.133342200178441, "no_speech_prob": 9.464740623188472e-12, "compression_ratio": 1.1474820143884892, "source": "whisper" }, { "start": 884.07, "end": 886.41, "text": "如果你有任何具体问题,请告诉我", "chunk": 1, "language": "zh", "avg_logprob": -0.133342200178441, "no_speech_prob": 9.464740623188472e-12, "compression_ratio": 1.1474820143884892, "source": "whisper" }, { "start": 887.05, "end": 888.13, "text": "无论如何", "chunk": 1, "language": "zh", "avg_logprob": -0.09224720490284455, "no_speech_prob": 1.125090385439842e-11, "compression_ratio": 0.9896907216494846, "source": "whisper" }, { "start": 888.13, "end": 889.79, "text": "我希望你觉得这个视频有用", "chunk": 1, "language": "zh", "avg_logprob": -0.09224720490284455, "no_speech_prob": 1.125090385439842e-11, "compression_ratio": 0.9896907216494846, "source": "whisper" }, { "start": 889.79, "end": 890.61, "text": "感谢观看", "chunk": 1, "language": "zh", "avg_logprob": -0.09224720490284455, "no_speech_prob": 1.125090385439842e-11, "compression_ratio": 0.9896907216494846, "source": "whisper" }, { "start": 890.61, "end": 891.47, "text": "像往常一样", "chunk": 1, "language": "zh", "avg_logprob": -0.09224720490284455, "no_speech_prob": 1.125090385439842e-11, "compression_ratio": 0.9896907216494846, "source": "whisper" }, { "start": 891.47, "end": 892.71, "text": "我们下个视频见", "chunk": 1, "language": "zh", "avg_logprob": -0.09224720490284455, "no_speech_prob": 1.125090385439842e-11, "compression_ratio": 0.9896907216494846, "source": "whisper" } ], "transcript_selection": { "schema": "videonote.transcript_selection.v1", "status": "SELECTED", "requested_language": "zh", "duration": 893.085938, "reference_path": null, "reference_kind": null, "reference_language": null, "selected_source": "whisper_segmented", "decision_reason": "source_unavailable_or_untrusted_whisper_passed", "whisper_rejected": false, "source_profile": { "segment_count": 0, "duration_coverage": 0.0, "usable_full": false, "usable_partial": false, "hallucination_region_count": 0, "hallucination_regions": [], "language": { "requested": "zh", "matched": false } }, "whisper_profile": { "segment_count": 304, "multilingual": false, "first_start": 0.0, "final_end": 892.71, "duration_coverage": 0.999579, "maximum_gap_seconds": 13.69, "inverted_count": 0, "language": { "requested": "zh", "matched": true, "latin": 300, "han": 2991, "kana": 0, "hangul": 0 }, "hallucination_region_count": 0, "hallucination_regions": [], "usable_full": true, "usable_partial": true }, "comparison": { "window_seconds": 30.0, "compared_window_count": 0, "median_score": 0.0, "divergent_window_count": 0, "divergent_ratio": 0.0, "divergent_windows": [] }, "timestamp_alignment": { "applied": false, "estimated_offset_seconds": 0.0, "matched_segment_count": 0, "match_scores": [] }, "caption_timing_repair": null, "filled_source_profile": null, "whisper_duration_guard": { "media_duration": 893.086, "rejected_segment_count": 0, "rejected_segments": [] } } }