{ "text": "仅凭模型后训练的优化\n模型性能就能暴涨超过30%\n一个推理激活的参数量只有13B的模型\n居然就逼近了Cloud Open 4.8模型的性能\n这就是Deepseek V4 Flash正式版\n时隔三个月\nDeepseek V4的正式版模型终于陆续登场\n就在7月31号下午\nV4 Flash的正式版率先亮相\n相比三个月前的阅览版\n正式版Flash模型保持的架构和模型参数不变\n仍然是284B总参数量\n13B激活参数的MOE模型\n但正式版重新进行了模型的后训练\n不仅Agent和编程的性能实现了跨越式的增长\n而且模型的响应格式也进行了全面升级\n在原先的OpenEye Compilation API的基础上\n进一步适配了OpenEye的Responses API\n并且官宣正式全面接入Codex的生态\n本期外科为你深度解读Deepseek V4 Flash模型特性和性能评测结果\n并为大家详细介绍V4模型\n接入全新的Parnest Agent API\nResponses API的使用方法\n以及接入Codex的完整流程\n帮助大家离门槛上手使用全新的Deepseek V4正式版模型\n同时国外科的文字版资料\nOpenEye Responses API的使用方法\n以及Codex接入的文字版教程\n都已经上限至复犯AI社区\n大家扫描频道的二维码即可免费领取\n首先是性能方面\n根据官方发布的评测结果\nDeepseek V4 Flash已经全面超越了GM5.2\n和Deepseek V4 Pro的预览版\n甚至在一些评分指标上已经逼近了OP4.8模型\n而根据实际的测试结果来看\n模型的阉城性能\nAgent性能\n以及对真实物理世界的理解能力都进步巨大\n甚至困扰Deepseek已久的\n鸭子骑自营车的SVG动画生成\n都被V4 Flash正式版轻松拿下\n简直和此前的V4预览版\n判若两个模型\n当然如果说性能进步是常规迭代\n那么本次V4正式版模型关心拥抱OpenAI的Responses API技术体系\n则预示着正式版模型将会是全面接容OpenAI\nCodex技术生态的原生Agent基座模型\n而这对于开发者来说\n无疑是一次巨大的第一层API开发范式的转变\n同时也预示着跋越即将发布的Deepseek Harness Agent\n将会和OpenAI的基础通信范式保持一致\n那么接下来就让我给大家简单解释下\n到底什么是Responses API\n以及其基础使用方法\n然后再详细介绍如何将Deepseek V4正式版模型接入Kodaks\n在过去很长长时间\nOpenAI制定的Chat Compilations通信标准\n是唯一跨模型统一的模型底层通信规范\n那么哪怕是前段时间刚上线不久的Kimi K3模型\n底层仍然采用的是Chat Compilations作为基础模型响应格式\n这其实是一种面向消息对列的高效编辑的一类API\n开发者可以非常高效地完成Messages的列表编辑\n并通过修改消息列表来引导模型完成任务\n但或许是因为消息对列的编辑\n只适合打造聊天机器人而不适合打造agent\n于是去年3月11号\nOpenAI推出了全新的全面适配agent任务的模型调用\n新版是Responses API\n而我当时也在第一时间对其进行了解读和评测\n不同于Chat Compilations API是面向对话来构建应用\nResponses API是一种面向任务来构建应用的更高级的API\n用户只需要描述任务关联工具\n说的题词词就能自动构建一个agent loop循环非常高效\n或者你也可以将其理解成Responses API就是OpenAI版的LongChain\n而现在DeepC V4正式版模型也是国内首个全面接入Responses API的模型\n而我们在第一时间对其整体性能表现进行了深度评测\n在海浪工具Skills和MCP调度情况下运行非常稳定\n同时我们在第一时间制作了DeepC Responses API的接入指南和使用教程\n并上线至复犯AI社区大扫码即可免费领取\n当然对于DeepC V4模型来说\n全面拥抱Responses API体系也就意味着可以不贫接入Codex的完整功能生态\n而官方给出的V4正式版接入Codex的流程也非常简单\n只需要一行命令即可同步切换Codex Cli,Codex App和ChainGBT的基座模型\n将其切换为DeepC V4正式版模型\n同时完成原始配置的备份并且可以随时复原\n以Windows系统为例\n完整切换基座模型的流程如下\n首先先打开PowerShell\n然后输入IRM命令\n从DeepC官方下载Codex配置文件修改脚本并自动运行\n然后根据提示进行模型选取或者恢复原始配置\n配置模型可以根据实际情况选择V4 Flash或者Pro正式版模型\n并根据提示设置API Key\n设置完成后\n只要看到Installation Complete的提示\n就说明配置成功\n而如果之后要切换模型或者恢复Codex的原始配置\n则再次运行相同脚本并进行相应选择即可\n然后打开HitGPT或者Codex应用\n看到左下方是DeepSeq\n并且模型是自定义\n就说明已经配置成功了\n接下来即可进行任意对话或者执行各类任务\n整个过程非常简单\n并且Codex Click也会同步发生修改\n而紧接着我们就在Codex命令行中\n对V4 Flash正式版模型进行一轮深度评测\n期间我们让V4 Flash围绕着一个3万多行多模态PDF检锁系统\n进行代码审查和结构优化\n一条指令即可完成整个Codex的\nPlan Task Subagent Web Search等各项功能的驱动\n并最终完成代码库的审核优化测试一整观者流程\n非常顺畅高效\n这次DeepSeq V4 Flash模型发布非常迅速\n在API全面上线同时\n模型权重也正式在Hugging Face上开源\n当然V4 Flash其实还只是开位赛\n还只是接下来V4 Pro正式版和DeepSeq版Cloud Code\nDeepSeq Harness Agent发布的前奏\n按照30%的性能涨幅预期\n我们有留相信\nV4 Pro正式版将会是全球第一款\n性能全面超越OPPO4.8的全新一代开源旗舰模型\n好了 以上就是本期视频的全部内容\n感谢大家的关注和三连支持\n更多DeepSeq V4正式版的学习资料\n技术文档 模型部署跟Agent开发教程等\n都已经上限至复办AI技术社区\n大家扫码即可领取\n五十九天 专注大家提供最扎实优质的技术内容\n我们下个视频 再见", "language": null, "duration": 368.709688, "segments": [ { "id": 0, "seek": 0, "start": 0.0, "end": 1.6400000000000001, "text": "仅凭模型后训练的优化", "tokens": [ 50365, 1550, 227, 6336, 255, 41908, 39823, 13547, 7422, 255, 10115, 225, 1546, 7384, 246, 23756, 50447 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 1, "seek": 0, "start": 1.6400000000000001, "end": 4.16, "text": "模型性能就能暴涨超过30%", "tokens": [ 50447, 41908, 39823, 21686, 8225, 3111, 8225, 23826, 112, 35681, 101, 19869, 16866, 3446, 4, 50573 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 2, "seek": 0, "start": 4.16, "end": 7.24, "text": "一个推理激活的参数量只有13B的模型", "tokens": [ 50573, 20182, 33597, 13876, 26373, 222, 25956, 1546, 2129, 224, 33188, 26748, 35244, 7668, 33, 1546, 41908, 39823, 50727 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 3, "seek": 0, "start": 7.24, "end": 9.98, "text": "居然就逼近了Cloud Open 4.8模型的性能", "tokens": [ 50727, 9636, 227, 5823, 3111, 2215, 120, 17463, 2289, 32787, 7238, 1017, 13, 23, 41908, 39823, 1546, 21686, 8225, 50864 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 4, "seek": 0, "start": 9.98, "end": 12.42, "text": "这就是Deepseek V4 Flash正式版", "tokens": [ 50864, 5562, 5620, 11089, 595, 405, 916, 691, 19, 20232, 15789, 27584, 42096, 50986 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 5, "seek": 0, "start": 12.42, "end": 13.38, "text": "时隔三个月", "tokens": [ 50986, 15729, 165, 1206, 10960, 7549, 6939, 51034 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 6, "seek": 0, "start": 13.38, "end": 16.4, "text": "Deepseek V4的正式版模型终于陆续登场", "tokens": [ 51034, 11089, 595, 405, 916, 691, 19, 1546, 15789, 27584, 42096, 41908, 39823, 10115, 230, 37732, 8842, 228, 10115, 255, 25874, 50255, 51185 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 7, "seek": 0, "start": 16.4, "end": 18.06, "text": "就在7月31号下午", "tokens": [ 51185, 3111, 3581, 22, 6939, 12967, 26987, 4438, 44237, 51268 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 8, "seek": 0, "start": 18.06, "end": 20.38, "text": "V4 Flash的正式版率先亮相", "tokens": [ 51268, 53, 19, 20232, 1546, 15789, 27584, 42096, 44866, 10108, 22991, 15106, 51384 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 9, "seek": 0, "start": 20.38, "end": 22.18, "text": "相比三个月前的阅览版", "tokens": [ 51384, 15106, 11706, 10960, 7549, 6939, 8945, 1546, 10034, 227, 6758, 230, 42096, 51474 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 10, "seek": 0, "start": 22.18, "end": 24.78, "text": "正式版Flash模型保持的架构和模型参数不变", "tokens": [ 51474, 15789, 27584, 42096, 25680, 1299, 41908, 39823, 24302, 17694, 1546, 7360, 114, 7360, 226, 12565, 41908, 39823, 2129, 224, 33188, 1960, 2129, 246, 51604 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 11, "seek": 0, "start": 24.78, "end": 27.060000000000002, "text": "仍然是284B总参数量", "tokens": [ 51604, 1550, 235, 5823, 1541, 11205, 19, 33, 33440, 2129, 224, 33188, 26748, 51718 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 12, "seek": 0, "start": 27.060000000000002, "end": 29.16, "text": "13B激活参数的MOE模型", "tokens": [ 51718, 7668, 33, 26373, 222, 25956, 2129, 224, 33188, 1546, 18976, 36, 41908, 39823, 51823 ], "temperature": 0, "avg_logprob": -0.11022486005510602, "compression_ratio": 1.363914373088685, "no_speech_prob": 2.3101987337015295e-11, "source": "whisper" }, { "id": 13, "seek": 2916, "start": 29.16, "end": 32.02, "text": "但正式版重新进行了模型的后训练", "tokens": [ 50365, 8395, 15789, 27584, 42096, 12624, 12560, 36700, 46606, 41908, 39823, 1546, 13547, 7422, 255, 10115, 225, 50508 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 14, "seek": 2916, "start": 32.02, "end": 35.4, "text": "不仅Agent和编程的性能实现了跨越式的增长", "tokens": [ 50508, 1960, 1550, 227, 32, 6930, 12565, 38109, 244, 29649, 1546, 21686, 8225, 24726, 20204, 2289, 6563, 101, 25761, 27584, 1546, 24228, 252, 32271, 50677 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 15, "seek": 2916, "start": 35.4, "end": 38.36, "text": "而且模型的响应格式也进行了全面升级", "tokens": [ 50677, 22942, 41908, 39823, 1546, 3858, 235, 44297, 30921, 27584, 6404, 36700, 46606, 11319, 8833, 41670, 16853, 100, 50825 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 16, "seek": 2916, "start": 38.36, "end": 40.94, "text": "在原先的OpenEye Compilation API的基础上", "tokens": [ 50825, 3581, 19683, 10108, 1546, 45569, 36, 1200, 6620, 16067, 9362, 1546, 26008, 38114, 222, 5708, 50954 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 17, "seek": 2916, "start": 40.94, "end": 43.22, "text": "进一步适配了OpenEye的Responses API", "tokens": [ 50954, 36700, 2257, 31429, 2215, 224, 38846, 2289, 45569, 36, 1200, 1546, 49, 13361, 892, 279, 9362, 51068 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 18, "seek": 2916, "start": 43.22, "end": 46.06, "text": "并且官宣正式全面接入Codex的生态", "tokens": [ 51068, 3509, 114, 20334, 31929, 2415, 96, 15789, 27584, 11319, 8833, 14468, 14028, 34, 1429, 87, 1546, 8244, 3757, 223, 51210 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 19, "seek": 2916, "start": 46.06, "end": 50.94, "text": "本期外科为你深度解读Deepseek V4 Flash模型特性和性能评测结果", "tokens": [ 51210, 8802, 16786, 12022, 42091, 13992, 2166, 24043, 13127, 17278, 5233, 119, 11089, 595, 405, 916, 691, 19, 20232, 41908, 39823, 17682, 21686, 12565, 21686, 8225, 5233, 226, 11038, 233, 45641, 9319, 51454 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 20, "seek": 2916, "start": 50.94, "end": 53.28, "text": "并为大家详细介绍V4模型", "tokens": [ 51454, 3509, 114, 13992, 6868, 5233, 99, 10115, 228, 30312, 10115, 235, 53, 19, 41908, 39823, 51571 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 21, "seek": 2916, "start": 53.28, "end": 55.379999999999995, "text": "接入全新的Parnest Agent API", "tokens": [ 51571, 14468, 14028, 11319, 12560, 1546, 47, 1083, 377, 27174, 9362, 51676 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 22, "seek": 2916, "start": 55.379999999999995, "end": 57.54, "text": "Responses API的使用方法", "tokens": [ 51676, 49, 13361, 892, 279, 9362, 1546, 22982, 9254, 9249, 11148, 51784 ], "temperature": 0, "avg_logprob": -0.11215065435035941, "compression_ratio": 1.2737430167597765, "no_speech_prob": 1.308055053161894e-11, "source": "whisper" }, { "id": 23, "seek": 5754, "start": 57.54, "end": 59.44, "text": "以及接入Codex的完整流程", "tokens": [ 50365, 40282, 14468, 14028, 34, 1429, 87, 1546, 14128, 27662, 27854, 29649, 50460 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 24, "seek": 5754, "start": 59.44, "end": 63.379999999999995, "text": "帮助大家离门槛上手使用全新的Deepseek V4正式版模型", "tokens": [ 50460, 4845, 106, 37618, 6868, 18614, 119, 8259, 101, 24880, 249, 5708, 11389, 22982, 9254, 11319, 12560, 1546, 11089, 595, 405, 916, 691, 19, 15789, 27584, 42096, 41908, 39823, 50657 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 25, "seek": 5754, "start": 63.379999999999995, "end": 65.26, "text": "同时国外科的文字版资料", "tokens": [ 50657, 13089, 15729, 16086, 12022, 42091, 1546, 17174, 22381, 42096, 5266, 226, 33404, 50751 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 26, "seek": 5754, "start": 65.26, "end": 67.5, "text": "OpenEye Responses API的使用方法", "tokens": [ 50751, 45569, 36, 1200, 46003, 279, 9362, 1546, 22982, 9254, 9249, 11148, 50863 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 27, "seek": 5754, "start": 67.5, "end": 69.8, "text": "以及Codex接入的文字版教程", "tokens": [ 50863, 40282, 34, 1429, 87, 14468, 14028, 1546, 17174, 22381, 42096, 21936, 29649, 50978 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 28, "seek": 5754, "start": 69.8, "end": 71.94, "text": "都已经上限至复犯AI社区", "tokens": [ 50978, 7182, 49161, 5708, 43446, 20844, 1787, 235, 35276, 107, 48698, 27658, 9937, 118, 51085 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 29, "seek": 5754, "start": 71.94, "end": 74.46000000000001, "text": "大家扫描频道的二维码即可免费领取", "tokens": [ 51085, 6868, 3416, 104, 11673, 237, 39752, 6025, 1546, 11217, 10115, 112, 23230, 223, 39127, 4429, 2347, 235, 18464, 117, 12501, 228, 29436, 51211 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 30, "seek": 5754, "start": 74.46000000000001, "end": 76.22, "text": "首先是性能方面", "tokens": [ 51211, 36490, 1541, 21686, 8225, 39604, 51299 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 31, "seek": 5754, "start": 76.22, "end": 78.44, "text": "根据官方发布的评测结果", "tokens": [ 51299, 31337, 26075, 106, 31929, 9249, 28926, 34688, 1546, 5233, 226, 11038, 233, 45641, 9319, 51410 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 32, "seek": 5754, "start": 78.44, "end": 81.62, "text": "Deepseek V4 Flash已经全面超越了GM5.2", "tokens": [ 51410, 11089, 595, 405, 916, 691, 19, 20232, 49161, 11319, 8833, 19869, 25761, 2289, 25152, 20, 13, 17, 51569 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 33, "seek": 5754, "start": 81.62, "end": 83.44, "text": "和Deepseek V4 Pro的预览版", "tokens": [ 51569, 12565, 11089, 595, 405, 916, 691, 19, 1705, 1546, 12501, 226, 6758, 230, 42096, 51660 ], "temperature": 0, "avg_logprob": -0.11286195640355512, "compression_ratio": 1.2052785923753666, "no_speech_prob": 1.9018423988437227e-11, "source": "whisper" }, { "id": 34, "seek": 8344, "start": 83.44, "end": 87.28, "text": "甚至在一些评分指标上已经逼近了OP4.8模型", "tokens": [ 50365, 17556, 20844, 3581, 38515, 5233, 226, 6627, 25922, 162, 3921, 5708, 49161, 2215, 120, 17463, 2289, 12059, 19, 13, 23, 41908, 39823, 50557 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 35, "seek": 8344, "start": 87.28, "end": 89.8, "text": "而根据实际的测试结果来看", "tokens": [ 50557, 11070, 31337, 26075, 106, 24726, 8842, 227, 1546, 11038, 233, 5233, 243, 45641, 9319, 6912, 4200, 50683 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 36, "seek": 8344, "start": 89.8, "end": 91.32, "text": "模型的阉城性能", "tokens": [ 50683, 41908, 39823, 1546, 10034, 231, 45277, 21686, 8225, 50759 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 37, "seek": 8344, "start": 91.32, "end": 92.2, "text": "Agent性能", "tokens": [ 50759, 32, 6930, 21686, 8225, 50803 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 38, "seek": 8344, "start": 92.2, "end": 95.86, "text": "以及对真实物理世界的理解能力都进步巨大", "tokens": [ 50803, 40282, 8713, 6303, 24726, 23516, 13876, 24486, 1546, 13876, 17278, 8225, 13486, 7182, 36700, 31429, 5238, 101, 3582, 50986 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 39, "seek": 8344, "start": 95.86, "end": 98.06, "text": "甚至困扰Deepseek已久的", "tokens": [ 50986, 17556, 20844, 3919, 108, 3416, 108, 11089, 595, 405, 916, 15701, 25320, 1546, 51096 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 40, "seek": 8344, "start": 98.06, "end": 100.68, "text": "鸭子骑自营车的SVG动画生成", "tokens": [ 51096, 165, 116, 255, 7626, 49657, 239, 9722, 10674, 98, 17819, 99, 1546, 50, 53, 38, 34961, 27126, 8244, 11336, 51227 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 41, "seek": 8344, "start": 100.68, "end": 103.08, "text": "都被V4 Flash正式版轻松拿下", "tokens": [ 51227, 7182, 23238, 53, 19, 20232, 15789, 27584, 42096, 17819, 119, 36897, 24351, 4438, 51347 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 42, "seek": 8344, "start": 103.08, "end": 105.5, "text": "简直和此前的V4预览版", "tokens": [ 51347, 11249, 222, 16186, 12565, 17947, 8945, 1546, 53, 19, 12501, 226, 6758, 230, 42096, 51468 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 43, "seek": 8344, "start": 105.5, "end": 106.82, "text": "判若两个模型", "tokens": [ 51468, 2437, 97, 13736, 98, 36257, 7549, 41908, 39823, 51534 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 44, "seek": 8344, "start": 106.82, "end": 110.02, "text": "当然如果说性能进步是常规迭代", "tokens": [ 51534, 40486, 13119, 8090, 21686, 8225, 36700, 31429, 1541, 11279, 6758, 226, 3316, 255, 19105, 51694 ], "temperature": 0, "avg_logprob": -0.1420759079773302, "compression_ratio": 1.1526479750778815, "no_speech_prob": 1.847844620372907e-11, "source": "whisper" }, { "id": 45, "seek": 11002, "start": 110.02, "end": 114.75999999999999, "text": "那么本次V4正式版模型关心拥抱OpenAI的Responses API技术体系", "tokens": [ 50365, 35693, 8802, 9487, 53, 19, 15789, 27584, 42096, 41908, 39823, 28053, 7945, 6852, 98, 38382, 45569, 48698, 1546, 49, 13361, 892, 279, 9362, 32502, 1474, 107, 29485, 25368, 50602 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 46, "seek": 11002, "start": 114.75999999999999, "end": 118.06, "text": "则预示着正式版模型将会是全面接容OpenAI", "tokens": [ 50602, 2437, 247, 12501, 226, 25696, 20708, 15789, 27584, 42096, 41908, 39823, 45456, 12949, 1541, 11319, 8833, 14468, 25750, 45569, 48698, 50767 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 47, "seek": 11002, "start": 118.06, "end": 121.53999999999999, "text": "Codex技术生态的原生Agent基座模型", "tokens": [ 50767, 34, 1429, 87, 32502, 1474, 107, 8244, 3757, 223, 1546, 19683, 8244, 32, 6930, 26008, 6346, 100, 41908, 39823, 50941 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 48, "seek": 11002, "start": 121.53999999999999, "end": 122.97999999999999, "text": "而这对于开发者来说", "tokens": [ 50941, 11070, 5562, 8713, 37732, 18937, 28926, 12444, 6912, 8090, 51013 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 49, "seek": 11002, "start": 122.97999999999999, "end": 126.14, "text": "无疑是一次巨大的第一层API开发范式的转变", "tokens": [ 51013, 31364, 18140, 239, 1541, 27505, 5238, 101, 39156, 18049, 9636, 224, 4715, 40, 18937, 28926, 27395, 225, 27584, 1546, 17819, 105, 2129, 246, 51171 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 50, "seek": 11002, "start": 126.14, "end": 130.22, "text": "同时也预示着跋越即将发布的Deepseek Harness Agent", "tokens": [ 51171, 13089, 15729, 6404, 12501, 226, 25696, 20708, 6563, 233, 25761, 39127, 45456, 28926, 34688, 1546, 11089, 595, 405, 916, 389, 16937, 27174, 51375 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 51, "seek": 11002, "start": 130.22, "end": 133.32, "text": "将会和OpenAI的基础通信范式保持一致", "tokens": [ 51375, 45456, 12949, 12565, 45569, 48698, 1546, 26008, 38114, 222, 19550, 17665, 27395, 225, 27584, 24302, 17694, 2257, 6784, 112, 51530 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 52, "seek": 11002, "start": 133.32, "end": 135.82, "text": "那么接下来就让我给大家简单解释下", "tokens": [ 51530, 35693, 14468, 4438, 6912, 3111, 33650, 1654, 23197, 6868, 11249, 222, 47446, 17278, 5873, 232, 4438, 51655 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 53, "seek": 11002, "start": 135.82, "end": 137.92, "text": "到底什么是Responses API", "tokens": [ 51655, 33883, 10440, 1541, 49, 13361, 892, 279, 9362, 51760 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 54, "seek": 11002, "start": 137.92, "end": 139.34, "text": "以及其基础使用方法", "tokens": [ 51760, 40282, 9572, 26008, 38114, 222, 22982, 9254, 9249, 11148, 51831 ], "temperature": 0, "avg_logprob": -0.11516429216433795, "compression_ratio": 1.292817679558011, "no_speech_prob": 1.7497064561111664e-11, "source": "whisper" }, { "id": 55, "seek": 13934, "start": 139.34, "end": 143.9, "text": "然后再详细介绍如何将Deepseek V4正式版模型接入Kodaks", "tokens": [ 50365, 26636, 8623, 5233, 99, 10115, 228, 30312, 10115, 235, 43526, 45456, 11089, 595, 405, 916, 691, 19, 15789, 27584, 42096, 41908, 39823, 14468, 14028, 42, 378, 5461, 50593 ], "temperature": 0, "avg_logprob": -0.098107258034818, "compression_ratio": 1.2372881355932204, "no_speech_prob": 1.525617653042044e-11, "source": "whisper" }, { "id": 56, "seek": 13934, "start": 143.9, "end": 145.28, "text": "在过去很长长时间", "tokens": [ 50593, 3581, 16866, 6734, 4563, 32271, 32271, 44848, 50662 ], "temperature": 0, "avg_logprob": -0.098107258034818, "compression_ratio": 1.2372881355932204, "no_speech_prob": 1.525617653042044e-11, "source": "whisper" }, { "id": 57, "seek": 13934, "start": 145.28, "end": 148.24, "text": "OpenAI制定的Chat Compilations通信标准", "tokens": [ 50662, 45569, 48698, 25491, 12088, 1546, 41683, 6620, 388, 763, 19550, 17665, 162, 3921, 6336, 228, 50810 ], "temperature": 0, "avg_logprob": -0.098107258034818, "compression_ratio": 1.2372881355932204, "no_speech_prob": 1.525617653042044e-11, "source": "whisper" }, { "id": 58, "seek": 13934, "start": 148.24, "end": 151.4, "text": "是唯一跨模型统一的模型底层通信规范", "tokens": [ 50810, 1541, 17198, 107, 2257, 6563, 101, 41908, 39823, 10115, 253, 2257, 1546, 41908, 39823, 22816, 9636, 224, 19550, 17665, 6758, 226, 27395, 225, 50968 ], "temperature": 0, "avg_logprob": -0.098107258034818, "compression_ratio": 1.2372881355932204, "no_speech_prob": 1.525617653042044e-11, "source": "whisper" }, { "id": 59, "seek": 13934, "start": 151.4, "end": 155.12, "text": "那么哪怕是前段时间刚上线不久的Kimi K3模型", "tokens": [ 50968, 35693, 17028, 21164, 1541, 8945, 28427, 44848, 49160, 5708, 16853, 123, 1960, 25320, 1546, 42, 10121, 591, 18, 41908, 39823, 51154 ], "temperature": 0, "avg_logprob": -0.098107258034818, "compression_ratio": 1.2372881355932204, "no_speech_prob": 1.525617653042044e-11, "source": "whisper" }, { "id": 60, "seek": 13934, "start": 155.12, "end": 159.06, "text": "底层仍然采用的是Chat Compilations作为基础模型响应格式", "tokens": [ 51154, 22816, 9636, 224, 1550, 235, 5823, 5873, 229, 9254, 24620, 41683, 6620, 388, 763, 11914, 13992, 26008, 38114, 222, 41908, 39823, 3858, 235, 44297, 30921, 27584, 51351 ], "temperature": 0, "avg_logprob": -0.098107258034818, "compression_ratio": 1.2372881355932204, "no_speech_prob": 1.525617653042044e-11, "source": "whisper" }, { "id": 61, "seek": 13934, "start": 159.06, "end": 162.92000000000002, "text": "这其实是一种面向消息对列的高效编辑的一类API", "tokens": [ 51351, 5562, 41646, 1541, 2257, 39810, 8833, 24282, 28837, 26460, 8713, 43338, 1546, 12979, 43076, 38109, 244, 9830, 239, 1546, 2257, 22113, 119, 4715, 40, 51544 ], "temperature": 0, "avg_logprob": -0.098107258034818, "compression_ratio": 1.2372881355932204, "no_speech_prob": 1.525617653042044e-11, "source": "whisper" }, { "id": 62, "seek": 13934, "start": 162.92000000000002, "end": 166.16, "text": "开发者可以非常高效地完成Messages的列表编辑", "tokens": [ 51544, 18937, 28926, 12444, 6723, 14392, 12979, 43076, 10928, 41509, 44, 442, 1660, 1546, 43338, 17571, 38109, 244, 9830, 239, 51706 ], "temperature": 0, "avg_logprob": -0.098107258034818, "compression_ratio": 1.2372881355932204, "no_speech_prob": 1.525617653042044e-11, "source": "whisper" }, { "id": 63, "seek": 16616, "start": 166.16, "end": 169.6, "text": "并通过修改消息列表来引导模型完成任务", "tokens": [ 50365, 3509, 114, 19550, 16866, 7792, 106, 34490, 28837, 26460, 43338, 17571, 6912, 41301, 4510, 120, 41908, 39823, 41509, 26443, 5087, 94, 50537 ], "temperature": 0, "avg_logprob": -0.09862810192686139, "compression_ratio": 1.1476014760147601, "no_speech_prob": 1.8919090252755844e-11, "source": "whisper" }, { "id": 64, "seek": 16616, "start": 169.6, "end": 172.04, "text": "但或许是因为消息对列的编辑", "tokens": [ 50537, 8395, 19780, 7422, 116, 1541, 34627, 28837, 26460, 8713, 43338, 1546, 38109, 244, 9830, 239, 50659 ], "temperature": 0, "avg_logprob": -0.09862810192686139, "compression_ratio": 1.1476014760147601, "no_speech_prob": 1.8919090252755844e-11, "source": "whisper" }, { "id": 65, "seek": 16616, "start": 172.04, "end": 175.1, "text": "只适合打造聊天机器人而不适合打造agent", "tokens": [ 50659, 14003, 2215, 224, 14245, 12467, 37583, 40096, 6135, 37960, 34386, 4035, 11070, 1960, 2215, 224, 14245, 12467, 37583, 559, 317, 50812 ], "temperature": 0, "avg_logprob": -0.09862810192686139, "compression_ratio": 1.1476014760147601, "no_speech_prob": 1.8919090252755844e-11, "source": "whisper" }, { "id": 66, "seek": 16616, "start": 175.1, "end": 176.56, "text": "于是去年3月11号", "tokens": [ 50812, 37732, 1541, 6734, 5157, 18, 6939, 5348, 26987, 50885 ], "temperature": 0, "avg_logprob": -0.09862810192686139, "compression_ratio": 1.1476014760147601, "no_speech_prob": 1.8919090252755844e-11, "source": "whisper" }, { "id": 67, "seek": 16616, "start": 176.56, "end": 181.04, "text": "OpenAI推出了全新的全面适配agent任务的模型调用", "tokens": [ 50885, 45569, 48698, 33597, 7781, 2289, 11319, 12560, 1546, 11319, 8833, 2215, 224, 38846, 559, 317, 26443, 5087, 94, 1546, 41908, 39823, 8897, 225, 9254, 51109 ], "temperature": 0, "avg_logprob": -0.09862810192686139, "compression_ratio": 1.1476014760147601, "no_speech_prob": 1.8919090252755844e-11, "source": "whisper" }, { "id": 68, "seek": 16616, "start": 181.04, "end": 182.68, "text": "新版是Responses API", "tokens": [ 51109, 12560, 42096, 1541, 49, 13361, 892, 279, 9362, 51191 ], "temperature": 0, "avg_logprob": -0.09862810192686139, "compression_ratio": 1.1476014760147601, "no_speech_prob": 1.8919090252755844e-11, "source": "whisper" }, { "id": 69, "seek": 16616, "start": 182.68, "end": 185.4, "text": "而我当时也在第一时间对其进行了解读和评测", "tokens": [ 51191, 11070, 1654, 16233, 15729, 6404, 3581, 18049, 44848, 8713, 9572, 36700, 46606, 17278, 5233, 119, 12565, 5233, 226, 11038, 233, 51327 ], "temperature": 0, "avg_logprob": -0.09862810192686139, "compression_ratio": 1.1476014760147601, "no_speech_prob": 1.8919090252755844e-11, "source": "whisper" }, { "id": 70, "seek": 18540, "start": 185.4, "end": 189.48000000000002, "text": "不同于Chat Compilations API是面向对话来构建应用", "tokens": [ 50365, 47123, 37732, 41683, 6620, 388, 763, 9362, 1541, 8833, 24282, 8713, 21596, 6912, 7360, 226, 34157, 44297, 9254, 50569 ], "temperature": 0, "avg_logprob": -0.12726181431820519, "compression_ratio": 1.2420382165605095, "no_speech_prob": 1.5628102978393343e-11, "source": "whisper" }, { "id": 71, "seek": 18540, "start": 189.48000000000002, "end": 194.28, "text": "Responses API是一种面向任务来构建应用的更高级的API", "tokens": [ 50569, 49, 13361, 892, 279, 9362, 1541, 2257, 39810, 8833, 24282, 26443, 5087, 94, 6912, 7360, 226, 34157, 44297, 9254, 1546, 19002, 12979, 16853, 100, 1546, 4715, 40, 50809 ], "temperature": 0, "avg_logprob": -0.12726181431820519, "compression_ratio": 1.2420382165605095, "no_speech_prob": 1.5628102978393343e-11, "source": "whisper" }, { "id": 72, "seek": 18540, "start": 194.28, "end": 196.98000000000002, "text": "用户只需要描述任务关联工具", "tokens": [ 50809, 9254, 1486, 115, 14003, 35748, 11673, 237, 3316, 108, 26443, 5087, 94, 28053, 8171, 242, 23323, 39806, 50944 ], "temperature": 0, "avg_logprob": -0.12726181431820519, "compression_ratio": 1.2420382165605095, "no_speech_prob": 1.5628102978393343e-11, "source": "whisper" }, { "id": 73, "seek": 18540, "start": 196.98000000000002, "end": 200.92000000000002, "text": "说的题词词就能自动构建一个agent loop循环非常高效", "tokens": [ 50944, 8090, 1546, 30716, 5233, 235, 5233, 235, 3111, 8225, 9722, 34961, 7360, 226, 34157, 20182, 559, 317, 6367, 2172, 103, 8051, 107, 14392, 12979, 43076, 51141 ], "temperature": 0, "avg_logprob": -0.12726181431820519, "compression_ratio": 1.2420382165605095, "no_speech_prob": 1.5628102978393343e-11, "source": "whisper" }, { "id": 74, "seek": 18540, "start": 200.92000000000002, "end": 205.08, "text": "或者你也可以将其理解成Responses API就是OpenAI版的LongChain", "tokens": [ 51141, 31148, 2166, 47935, 45456, 9572, 13876, 17278, 11336, 49, 13361, 892, 279, 9362, 5620, 45569, 48698, 42096, 1546, 43, 556, 6546, 491, 51349 ], "temperature": 0, "avg_logprob": -0.12726181431820519, "compression_ratio": 1.2420382165605095, "no_speech_prob": 1.5628102978393343e-11, "source": "whisper" }, { "id": 75, "seek": 18540, "start": 205.08, "end": 210.22, "text": "而现在DeepC V4正式版模型也是国内首个全面接入Responses API的模型", "tokens": [ 51349, 11070, 25040, 11089, 595, 34, 691, 19, 15789, 27584, 42096, 41908, 39823, 22021, 16086, 34742, 25444, 7549, 11319, 8833, 14468, 14028, 49, 13361, 892, 279, 9362, 1546, 41908, 39823, 51606 ], "temperature": 0, "avg_logprob": -0.12726181431820519, "compression_ratio": 1.2420382165605095, "no_speech_prob": 1.5628102978393343e-11, "source": "whisper" }, { "id": 76, "seek": 21022, "start": 210.22, "end": 214.04, "text": "而我们在第一时间对其整体性能表现进行了深度评测", "tokens": [ 50365, 11070, 15003, 3581, 18049, 44848, 8713, 9572, 27662, 29485, 21686, 8225, 17571, 20204, 36700, 46606, 24043, 13127, 5233, 226, 11038, 233, 50556 ], "temperature": 0, "avg_logprob": -0.10938097979571368, "compression_ratio": 1.2583333333333333, "no_speech_prob": 1.4740366913179592e-11, "source": "whisper" }, { "id": 77, "seek": 21022, "start": 214.04, "end": 217.92, "text": "在海浪工具Skills和MCP调度情况下运行非常稳定", "tokens": [ 50556, 3581, 20078, 11038, 103, 23323, 39806, 50, 74, 2565, 12565, 44, 20049, 8897, 225, 13127, 46514, 4438, 3316, 238, 8082, 14392, 10415, 111, 12088, 50750 ], "temperature": 0, "avg_logprob": -0.10938097979571368, "compression_ratio": 1.2583333333333333, "no_speech_prob": 1.4740366913179592e-11, "source": "whisper" }, { "id": 78, "seek": 21022, "start": 217.92, "end": 222.5, "text": "同时我们在第一时间制作了DeepC Responses API的接入指南和使用教程", "tokens": [ 50750, 13089, 15729, 15003, 3581, 18049, 44848, 25491, 11914, 2289, 11089, 595, 34, 22480, 892, 279, 9362, 1546, 14468, 14028, 25922, 31021, 12565, 22982, 9254, 21936, 29649, 50979 ], "temperature": 0, "avg_logprob": -0.10938097979571368, "compression_ratio": 1.2583333333333333, "no_speech_prob": 1.4740366913179592e-11, "source": "whisper" }, { "id": 79, "seek": 21022, "start": 222.5, "end": 225.52, "text": "并上线至复犯AI社区大扫码即可免费领取", "tokens": [ 50979, 3509, 114, 5708, 16853, 123, 20844, 1787, 235, 35276, 107, 48698, 27658, 9937, 118, 3582, 3416, 104, 23230, 223, 39127, 4429, 2347, 235, 18464, 117, 12501, 228, 29436, 51130 ], "temperature": 0, "avg_logprob": -0.10938097979571368, "compression_ratio": 1.2583333333333333, "no_speech_prob": 1.4740366913179592e-11, "source": "whisper" }, { "id": 80, "seek": 21022, "start": 225.52, "end": 227.92, "text": "当然对于DeepC V4模型来说", "tokens": [ 51130, 40486, 8713, 37732, 11089, 595, 34, 691, 19, 41908, 39823, 6912, 8090, 51250 ], "temperature": 0, "avg_logprob": -0.10938097979571368, "compression_ratio": 1.2583333333333333, "no_speech_prob": 1.4740366913179592e-11, "source": "whisper" }, { "id": 81, "seek": 21022, "start": 227.92, "end": 232.84, "text": "全面拥抱Responses API体系也就意味着可以不贫接入Codex的完整功能生态", "tokens": [ 51250, 11319, 8833, 6852, 98, 38382, 49, 13361, 892, 279, 9362, 29485, 25368, 6404, 3111, 9042, 17268, 20708, 6723, 1960, 18464, 104, 14468, 14028, 34, 1429, 87, 1546, 14128, 27662, 32311, 8225, 8244, 3757, 223, 51496 ], "temperature": 0, "avg_logprob": -0.10938097979571368, "compression_ratio": 1.2583333333333333, "no_speech_prob": 1.4740366913179592e-11, "source": "whisper" }, { "id": 82, "seek": 21022, "start": 232.84, "end": 237.07999999999998, "text": "而官方给出的V4正式版接入Codex的流程也非常简单", "tokens": [ 51496, 11070, 31929, 9249, 23197, 7781, 1546, 53, 19, 15789, 27584, 42096, 14468, 14028, 34, 1429, 87, 1546, 27854, 29649, 6404, 14392, 11249, 222, 47446, 51708 ], "temperature": 0, "avg_logprob": -0.10938097979571368, "compression_ratio": 1.2583333333333333, "no_speech_prob": 1.4740366913179592e-11, "source": "whisper" }, { "id": 83, "seek": 23708, "start": 237.08, "end": 242.72, "text": "只需要一行命令即可同步切换Codex Cli,Codex App和ChainGBT的基座模型", "tokens": [ 50365, 14003, 35748, 2257, 8082, 25236, 49061, 39127, 4429, 13089, 31429, 23632, 26075, 95, 34, 1429, 87, 2033, 72, 11, 34, 1429, 87, 3132, 12565, 6546, 491, 8769, 51, 1546, 26008, 6346, 100, 41908, 39823, 50647 ], "temperature": 0, "avg_logprob": -0.10242350896199544, "compression_ratio": 1.2238805970149254, "no_speech_prob": 1.5034017433745994e-11, "source": "whisper" }, { "id": 84, "seek": 23708, "start": 242.72, "end": 245.02, "text": "将其切换为DeepC V4正式版模型", "tokens": [ 50647, 45456, 9572, 23632, 26075, 95, 13992, 11089, 595, 34, 691, 19, 15789, 27584, 42096, 41908, 39823, 50762 ], "temperature": 0, "avg_logprob": -0.10242350896199544, "compression_ratio": 1.2238805970149254, "no_speech_prob": 1.5034017433745994e-11, "source": "whisper" }, { "id": 85, "seek": 23708, "start": 245.02, "end": 248.92000000000002, "text": "同时完成原始配置的备份并且可以随时复原", "tokens": [ 50762, 13089, 15729, 41509, 19683, 15476, 38846, 34719, 1546, 1787, 229, 36266, 3509, 114, 20334, 6723, 10673, 237, 15729, 1787, 235, 19683, 50957 ], "temperature": 0, "avg_logprob": -0.10242350896199544, "compression_ratio": 1.2238805970149254, "no_speech_prob": 1.5034017433745994e-11, "source": "whisper" }, { "id": 86, "seek": 23708, "start": 248.92000000000002, "end": 250.36, "text": "以Windows系统为例", "tokens": [ 50957, 3588, 45813, 1509, 25368, 10115, 253, 13992, 17797, 51029 ], "temperature": 0, "avg_logprob": -0.10242350896199544, "compression_ratio": 1.2238805970149254, "no_speech_prob": 1.5034017433745994e-11, "source": "whisper" }, { "id": 87, "seek": 23708, "start": 250.36, "end": 252.5, "text": "完整切换基座模型的流程如下", "tokens": [ 51029, 14128, 27662, 23632, 26075, 95, 26008, 6346, 100, 41908, 39823, 1546, 27854, 29649, 8238, 4438, 51136 ], "temperature": 0, "avg_logprob": -0.10242350896199544, "compression_ratio": 1.2238805970149254, "no_speech_prob": 1.5034017433745994e-11, "source": "whisper" }, { "id": 88, "seek": 23708, "start": 252.5, "end": 254.26000000000002, "text": "首先先打开PowerShell", "tokens": [ 51136, 36490, 10108, 12467, 18937, 46057, 9526, 285, 51224 ], "temperature": 0, "avg_logprob": -0.10242350896199544, "compression_ratio": 1.2238805970149254, "no_speech_prob": 1.5034017433745994e-11, "source": "whisper" }, { "id": 89, "seek": 23708, "start": 254.26000000000002, "end": 256.54, "text": "然后输入IRM命令", "tokens": [ 51224, 26636, 9830, 241, 14028, 7740, 44, 25236, 49061, 51338 ], "temperature": 0, "avg_logprob": -0.10242350896199544, "compression_ratio": 1.2238805970149254, "no_speech_prob": 1.5034017433745994e-11, "source": "whisper" }, { "id": 90, "seek": 23708, "start": 256.54, "end": 260.88, "text": "从DeepC官方下载Codex配置文件修改脚本并自动运行", "tokens": [ 51338, 35630, 11089, 595, 34, 31929, 9249, 4438, 17819, 121, 34, 1429, 87, 38846, 34719, 17174, 20485, 7792, 106, 34490, 27067, 248, 8802, 3509, 114, 9722, 34961, 3316, 238, 8082, 51555 ], "temperature": 0, "avg_logprob": -0.10242350896199544, "compression_ratio": 1.2238805970149254, "no_speech_prob": 1.5034017433745994e-11, "source": "whisper" }, { "id": 91, "seek": 23708, "start": 260.88, "end": 265.24, "text": "然后根据提示进行模型选取或者恢复原始配置", "tokens": [ 51555, 26636, 31337, 26075, 106, 20949, 25696, 36700, 8082, 41908, 39823, 2215, 231, 29436, 31148, 11845, 95, 1787, 235, 19683, 15476, 38846, 34719, 51773 ], "temperature": 0, "avg_logprob": -0.10242350896199544, "compression_ratio": 1.2238805970149254, "no_speech_prob": 1.5034017433745994e-11, "source": "whisper" }, { "id": 92, "seek": 26524, "start": 265.24, "end": 269.54, "text": "配置模型可以根据实际情况选择V4 Flash或者Pro正式版模型", "tokens": [ 50365, 38846, 34719, 41908, 39823, 6723, 31337, 26075, 106, 24726, 8842, 227, 46514, 2215, 231, 6852, 102, 53, 19, 20232, 31148, 12681, 15789, 27584, 42096, 41908, 39823, 50580 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 93, "seek": 26524, "start": 269.54, "end": 271.72, "text": "并根据提示设置API Key", "tokens": [ 50580, 3509, 114, 31337, 26075, 106, 20949, 25696, 7422, 122, 34719, 4715, 40, 12759, 50689 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 94, "seek": 26524, "start": 271.72, "end": 272.64, "text": "设置完成后", "tokens": [ 50689, 7422, 122, 34719, 41509, 13547, 50735 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 95, "seek": 26524, "start": 272.64, "end": 275.42, "text": "只要看到Installation Complete的提示", "tokens": [ 50735, 14003, 4275, 18032, 28411, 336, 399, 34687, 1546, 20949, 25696, 50874 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 96, "seek": 26524, "start": 275.42, "end": 276.88, "text": "就说明配置成功", "tokens": [ 50874, 3111, 8090, 11100, 38846, 34719, 11336, 32311, 50947 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 97, "seek": 26524, "start": 276.88, "end": 280.86, "text": "而如果之后要切换模型或者恢复Codex的原始配置", "tokens": [ 50947, 11070, 13119, 9574, 13547, 4275, 23632, 26075, 95, 41908, 39823, 31148, 11845, 95, 1787, 235, 34, 1429, 87, 1546, 19683, 15476, 38846, 34719, 51146 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 98, "seek": 26524, "start": 280.86, "end": 284.72, "text": "则再次运行相同脚本并进行相应选择即可", "tokens": [ 51146, 2437, 247, 8623, 9487, 3316, 238, 8082, 15106, 13089, 27067, 248, 8802, 3509, 114, 36700, 8082, 15106, 44297, 2215, 231, 6852, 102, 39127, 4429, 51339 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 99, "seek": 26524, "start": 284.72, "end": 287.7, "text": "然后打开HitGPT或者Codex应用", "tokens": [ 51339, 26636, 12467, 18937, 39, 270, 38, 47, 51, 31148, 34, 1429, 87, 44297, 9254, 51488 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 100, "seek": 26524, "start": 287.7, "end": 289.5, "text": "看到左下方是DeepSeq", "tokens": [ 51488, 18032, 21975, 4438, 9249, 1541, 11089, 595, 10637, 80, 51578 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 101, "seek": 26524, "start": 289.5, "end": 291.26, "text": "并且模型是自定义", "tokens": [ 51578, 3509, 114, 20334, 41908, 39823, 1541, 9722, 12088, 2930, 231, 51666 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 102, "seek": 26524, "start": 291.26, "end": 292.96000000000004, "text": "就说明已经配置成功了", "tokens": [ 51666, 3111, 8090, 11100, 49161, 38846, 34719, 11336, 32311, 2289, 51751 ], "temperature": 0, "avg_logprob": -0.10719123928026221, "compression_ratio": 1.2576687116564418, "no_speech_prob": 1.6632127966542676e-11, "source": "whisper" }, { "id": 103, "seek": 29296, "start": 292.96, "end": 296.4, "text": "接下来即可进行任意对话或者执行各类任务", "tokens": [ 50365, 14468, 4438, 6912, 39127, 4429, 36700, 8082, 26443, 9042, 8713, 21596, 31148, 3416, 100, 8082, 17516, 22113, 119, 26443, 5087, 94, 50537 ], "temperature": 0, "avg_logprob": -0.1274691169912165, "compression_ratio": 1.1820809248554913, "no_speech_prob": 1.7585325556845888e-11, "source": "whisper" }, { "id": 104, "seek": 29296, "start": 296.4, "end": 297.5, "text": "整个过程非常简单", "tokens": [ 50537, 27662, 7549, 16866, 29649, 14392, 11249, 222, 47446, 50592 ], "temperature": 0, "avg_logprob": -0.1274691169912165, "compression_ratio": 1.1820809248554913, "no_speech_prob": 1.7585325556845888e-11, "source": "whisper" }, { "id": 105, "seek": 29296, "start": 297.5, "end": 300.09999999999997, "text": "并且Codex Click也会同步发生修改", "tokens": [ 50592, 3509, 114, 20334, 34, 1429, 87, 8230, 6404, 12949, 13089, 31429, 28926, 8244, 7792, 106, 34490, 50722 ], "temperature": 0, "avg_logprob": -0.1274691169912165, "compression_ratio": 1.1820809248554913, "no_speech_prob": 1.7585325556845888e-11, "source": "whisper" }, { "id": 106, "seek": 29296, "start": 300.09999999999997, "end": 302.7, "text": "而紧接着我们就在Codex命令行中", "tokens": [ 50722, 11070, 7732, 100, 14468, 20708, 15003, 3111, 3581, 34, 1429, 87, 25236, 49061, 8082, 5975, 50852 ], "temperature": 0, "avg_logprob": -0.1274691169912165, "compression_ratio": 1.1820809248554913, "no_speech_prob": 1.7585325556845888e-11, "source": "whisper" }, { "id": 107, "seek": 29296, "start": 302.7, "end": 305.44, "text": "对V4 Flash正式版模型进行一轮深度评测", "tokens": [ 50852, 8713, 53, 19, 20232, 15789, 27584, 42096, 41908, 39823, 36700, 8082, 2257, 17819, 106, 24043, 13127, 5233, 226, 11038, 233, 50989 ], "temperature": 0, "avg_logprob": -0.1274691169912165, "compression_ratio": 1.1820809248554913, "no_speech_prob": 1.7585325556845888e-11, "source": "whisper" }, { "id": 108, "seek": 29296, "start": 305.44, "end": 310.08, "text": "期间我们让V4 Flash围绕着一个3万多行多模态PDF检锁系统", "tokens": [ 50989, 16786, 31685, 15003, 33650, 53, 19, 20232, 3919, 112, 10115, 243, 20708, 20182, 18, 23570, 6392, 8082, 6392, 41908, 3757, 223, 17349, 37, 16407, 222, 23049, 223, 25368, 10115, 253, 51221 ], "temperature": 0, "avg_logprob": -0.1274691169912165, "compression_ratio": 1.1820809248554913, "no_speech_prob": 1.7585325556845888e-11, "source": "whisper" }, { "id": 109, "seek": 29296, "start": 310.08, "end": 312.02, "text": "进行代码审查和结构优化", "tokens": [ 51221, 36700, 8082, 19105, 23230, 223, 2415, 94, 42623, 12565, 45641, 7360, 226, 7384, 246, 23756, 51318 ], "temperature": 0, "avg_logprob": -0.1274691169912165, "compression_ratio": 1.1820809248554913, "no_speech_prob": 1.7585325556845888e-11, "source": "whisper" }, { "id": 110, "seek": 29296, "start": 312.02, "end": 315.53999999999996, "text": "一条指令即可完成整个Codex的", "tokens": [ 51318, 2257, 48837, 25922, 49061, 39127, 4429, 41509, 27662, 7549, 34, 1429, 87, 1546, 51494 ], "temperature": 0, "avg_logprob": -0.1274691169912165, "compression_ratio": 1.1820809248554913, "no_speech_prob": 1.7585325556845888e-11, "source": "whisper" }, { "id": 111, "seek": 29296, "start": 315.53999999999996, "end": 318.96, "text": "Plan Task Subagent Web Search等各项功能的驱动", "tokens": [ 51494, 47, 8658, 30428, 8511, 559, 317, 9573, 17180, 10187, 17516, 10178, 117, 32311, 8225, 1546, 24023, 109, 34961, 51665 ], "temperature": 0, "avg_logprob": -0.1274691169912165, "compression_ratio": 1.1820809248554913, "no_speech_prob": 1.7585325556845888e-11, "source": "whisper" }, { "id": 112, "seek": 31896, "start": 318.96, "end": 322.96, "text": "并最终完成代码库的审核优化测试一整观者流程", "tokens": [ 50365, 3509, 114, 8661, 10115, 230, 41509, 19105, 23230, 223, 6346, 241, 1546, 2415, 94, 8987, 116, 7384, 246, 23756, 11038, 233, 5233, 243, 2257, 27662, 42350, 12444, 27854, 29649, 50565 ], "temperature": 0, "avg_logprob": -0.15528564453125, "compression_ratio": 1.147887323943662, "no_speech_prob": 1.8517224212311056e-11, "source": "whisper" }, { "id": 113, "seek": 31896, "start": 322.96, "end": 324.4, "text": "非常顺畅高效", "tokens": [ 50565, 14392, 10178, 118, 6904, 227, 12979, 43076, 50637 ], "temperature": 0, "avg_logprob": -0.15528564453125, "compression_ratio": 1.147887323943662, "no_speech_prob": 1.8517224212311056e-11, "source": "whisper" }, { "id": 114, "seek": 31896, "start": 324.4, "end": 326.79999999999995, "text": "这次DeepSeq V4 Flash模型发布非常迅速", "tokens": [ 50637, 5562, 9487, 11089, 595, 10637, 80, 691, 19, 20232, 41908, 39823, 28926, 34688, 14392, 3316, 227, 31217, 50757 ], "temperature": 0, "avg_logprob": -0.15528564453125, "compression_ratio": 1.147887323943662, "no_speech_prob": 1.8517224212311056e-11, "source": "whisper" }, { "id": 115, "seek": 31896, "start": 326.79999999999995, "end": 328.79999999999995, "text": "在API全面上线同时", "tokens": [ 50757, 3581, 4715, 40, 11319, 8833, 5708, 16853, 123, 13089, 15729, 50857 ], "temperature": 0, "avg_logprob": -0.15528564453125, "compression_ratio": 1.147887323943662, "no_speech_prob": 1.8517224212311056e-11, "source": "whisper" }, { "id": 116, "seek": 31896, "start": 328.79999999999995, "end": 331.34, "text": "模型权重也正式在Hugging Face上开源", "tokens": [ 50857, 41908, 39823, 4422, 225, 12624, 6404, 15789, 27584, 3581, 39, 697, 3249, 4047, 5708, 18937, 47402, 50984 ], "temperature": 0, "avg_logprob": -0.15528564453125, "compression_ratio": 1.147887323943662, "no_speech_prob": 1.8517224212311056e-11, "source": "whisper" }, { "id": 117, "seek": 31896, "start": 331.34, "end": 334.08, "text": "当然V4 Flash其实还只是开位赛", "tokens": [ 50984, 40486, 53, 19, 20232, 41646, 14852, 36859, 18937, 11160, 5266, 249, 51121 ], "temperature": 0, "avg_logprob": -0.15528564453125, "compression_ratio": 1.147887323943662, "no_speech_prob": 1.8517224212311056e-11, "source": "whisper" }, { "id": 118, "seek": 31896, "start": 334.08, "end": 337.84, "text": "还只是接下来V4 Pro正式版和DeepSeq版Cloud Code", "tokens": [ 51121, 14852, 36859, 14468, 4438, 6912, 53, 19, 1705, 15789, 27584, 42096, 12565, 11089, 595, 10637, 80, 42096, 32787, 15549, 51309 ], "temperature": 0, "avg_logprob": -0.15528564453125, "compression_ratio": 1.147887323943662, "no_speech_prob": 1.8517224212311056e-11, "source": "whisper" }, { "id": 119, "seek": 31896, "start": 337.84, "end": 340.82, "text": "DeepSeq Harness Agent发布的前奏", "tokens": [ 51309, 11089, 595, 10637, 80, 389, 16937, 27174, 28926, 34688, 1546, 8945, 1881, 237, 51458 ], "temperature": 0, "avg_logprob": -0.15528564453125, "compression_ratio": 1.147887323943662, "no_speech_prob": 1.8517224212311056e-11, "source": "whisper" }, { "id": 120, "seek": 34082, "start": 340.82, "end": 343.82, "text": "按照30%的性能涨幅预期", "tokens": [ 50365, 26613, 32150, 3446, 4, 1546, 21686, 8225, 35681, 101, 3509, 227, 12501, 226, 16786, 50515 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 121, "seek": 34082, "start": 343.82, "end": 344.7, "text": "我们有留相信", "tokens": [ 50515, 15003, 2412, 24456, 35428, 50559 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 122, "seek": 34082, "start": 344.7, "end": 347.26, "text": "V4 Pro正式版将会是全球第一款", "tokens": [ 50559, 53, 19, 1705, 15789, 27584, 42096, 45456, 12949, 1541, 11319, 28533, 18049, 48798, 50687 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 123, "seek": 34082, "start": 347.26, "end": 351.2, "text": "性能全面超越OPPO4.8的全新一代开源旗舰模型", "tokens": [ 50687, 21686, 8225, 11319, 8833, 19869, 25761, 12059, 34885, 19, 13, 23, 1546, 11319, 12560, 2257, 19105, 18937, 47402, 4479, 245, 10256, 108, 41908, 39823, 50884 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 124, "seek": 34082, "start": 351.2, "end": 353.15999999999997, "text": "好了 以上就是本期视频的全部内容", "tokens": [ 50884, 12621, 220, 29497, 5620, 8802, 16786, 40656, 39752, 1546, 38714, 34742, 25750, 50982 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 125, "seek": 34082, "start": 353.15999999999997, "end": 354.74, "text": "感谢大家的关注和三连支持", "tokens": [ 50982, 9709, 11340, 6868, 1546, 28053, 26432, 12565, 10960, 3316, 252, 35488, 51061 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 126, "seek": 34082, "start": 354.74, "end": 356.84, "text": "更多DeepSeq V4正式版的学习资料", "tokens": [ 51061, 19002, 6392, 11089, 595, 10637, 80, 691, 19, 15789, 27584, 42096, 1546, 29618, 2930, 254, 5266, 226, 33404, 51166 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 127, "seek": 34082, "start": 356.84, "end": 359.7, "text": "技术文档 模型部署跟Agent开发教程等", "tokens": [ 51166, 32502, 1474, 107, 17174, 21416, 96, 220, 41908, 39823, 13470, 16469, 110, 9678, 32, 6930, 18937, 28926, 21936, 29649, 10187, 51309 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 128, "seek": 34082, "start": 359.7, "end": 361.74, "text": "都已经上限至复办AI技术社区", "tokens": [ 51309, 7182, 49161, 5708, 43446, 20844, 1787, 235, 39453, 48698, 32502, 1474, 107, 27658, 9937, 118, 51411 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 129, "seek": 34082, "start": 361.74, "end": 363.1, "text": "大家扫码即可领取", "tokens": [ 51411, 6868, 3416, 104, 23230, 223, 39127, 4429, 12501, 228, 29436, 51479 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 130, "seek": 34082, "start": 363.1, "end": 366.68, "text": "五十九天 专注大家提供最扎实优质的技术内容", "tokens": [ 51479, 21001, 20145, 43145, 6135, 220, 940, 241, 26432, 6868, 20949, 3254, 249, 8661, 3416, 236, 24726, 7384, 246, 18464, 101, 1546, 32502, 1474, 107, 34742, 25750, 51658 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" }, { "id": 131, "seek": 34082, "start": 366.68, "end": 368.28, "text": "我们下个视频 再见", "tokens": [ 51658, 15003, 4438, 7549, 40656, 39752, 220, 44176, 51738 ], "temperature": 0, "avg_logprob": -0.13400947149075457, "compression_ratio": 1.2281167108753315, "no_speech_prob": 1.557325449152991e-11, "source": "whisper" } ], "transcript_selection": { "schema": "videonote.transcript_selection.v1", "status": "SELECTED", "requested_language": null, "duration": 368.709688, "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": null, "matched": false } }, "whisper_profile": { "segment_count": 132, "multilingual": false, "first_start": 0.0, "final_end": 368.28, "duration_coverage": 0.998835, "maximum_gap_seconds": 0.0, "inverted_count": 0, "language": { "requested": null, "matched": true, "latin": 825, "han": 1603, "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": [] }, "filled_source_profile": null, "whisper_duration_guard": { "media_duration": 368.71, "rejected_segment_count": 0, "rejected_segments": [] } } }