{"text": "Hello 今天咱们来聊点超有意思的AI圈新鲜事绝对能刷新你之前对大模型的刻板印象今天我们来聊聊一个有点意思的模型Mindforge 27B听到名字里的27B你可能会觉得这不就是个小参数模型吗毕竟现在动辄千亿参数的AI模型满天飞27B顶多算个小个子但就是这个小个子在编程能力上却干翻了不少大块头在编程测试基准Program Bench上它的第一次尝试通过率Pasit-E达到了49.51%比DeepSeek第四代专业版的47.80%还高甚至逼近了Cloud Opus 4.7的51.38%要知道Cloud Opus 4.7可是公认的编程高手而Mindforge 27B只用了1001条数据训练这到底是咋做到的咱们今天就来扒一扒它背后的技术细节首先得说这个模型的核心不在参数多而在数据精传统的大模型训练编程能力基本都是海量堆数据GitHub上几百万个项目几十亿行代码片段甚至包括各种开源库的函数定义注释用法视力一股脑往模型里塞这种训练方式确实能让模型学会写代码的语法比如怎么定义变量写循环掉裤但有个大问题它学到的是局部技能而不是工程思维就像你让一个只会被单词的人写小说他可能词都认识但写不出连贯的故事Mindforge 27B完全反其道而行之它没用海量代码片段而是用了1001条完整开发轨迹啥叫完整开发轨迹咱们打个比方你让一个程序员开发一个电商订单管理系统传统数据可能只给他订单创建这个函数的代码片段或者库存检查的一段逻辑但完整开发轨迹是从老板提需求开始到程序员分析需求设计系统架构画流程图写代码测bug改bug最后上线整个过程每一步的对话思考代码修改错误反馈全都有记录具体来说这1001条轨迹平均每条有181.6轮对话这不是简单的问答而是像真实开发团队里的协作比如第一轮用户也就是需求方说我们需要一个订单系统要支持多种支付方式还要能实时查库存模型作为开发者不会直接写代码而是先问细节支付方式需要支持支付宝微信还是银行卡实时查库存是扣库存钱查还是扣库存实查订单取消后库存要不要回退用户回答后模型会接着设计架构我们可以分订单服务支付服务库存服务三个模块用消息对列结偶然后写代码的时候模型会写订单表的创建语句写接口定义写业务逻辑写完跑测试比如发现高并发时库存扣减不对模型会根据错误日志分析原因应该是没加分布是锁导致两个请求同时读到库存为一都扣了然后修改代码加锁再测试直到通过你看这181.6轮对话里每一轮都包含三类信息上下文之前的需求讨论架构设计动作写了什么代码改了哪里反馈测试报了什么错用户提了什么意见模型训练的时候就是学习这三者之间的因果链根据上下文应该做什么动作做了动作会得到什么反馈得到反馈后该怎么调整这比单纯学代码片段长什么样高明太多了他学到的是开发代码的全流程逻辑也就是工程思维那这些轨迹数据是怎么来的研究团队从开源项目的完整开发历史里挖出来的比如Github上一些知名项目的issue讨论区PolRequest的修改记录代码评审的评论甚至是开发者之间的聊天记录当然脱敏了他们把这些非结构化的数据整理成结构化的对话代码反馈链每条轨迹都像一部开发日志记录了一个功能从无到有的全过程比如某个轨迹可能是开发一个用户登陆健全功能从最开始讨论用JWT还是Session到设计Token刷新机制到写登陆接口代码到测试发现Token过期时间太短再到调整过期时间加刷新接口最后上线这些细节传统代码数据里根本不会保留接下来是训练策略研究团队用了英国语言建模但不是普通的英国建模普通建模是把一段文本拆成Token让模型预测下一个Token而MindForge 27B把轨迹拆成片段每个片段可能是用户需求模型回复代码快错误日志然后让模型学习在什么上下文下该生成什么类型的片段比如前面是用户提了一个bug反馈模型就该预测错误分析和代码修改片段而不是直接生成代码片段这种结构化预测让模型更理解开发流程的节奏还有一个关键点模型在训练时会模拟错误传统训练里代码数据大多是正确的比如开源项目的最终代码模型很少看到错误代码和修复过程但MindForge 27B的轨迹里包含大量错误修复的循环比如模型写了一段代码测试爆空指针异常然后模型会分析哪个变量可能为空修改代码加判空再测试可能又爆类型不匹配再调整这种试错过程让模型学会了如何调试这不是靠记住常见bug而是掌握了排查问题的思路就像一个老程序员不是背会了所有bug的解法而是掌握了排查问题的思路那效果为啥这么好咱们再看实验细节ProgramBench是一个很严格的编程测试集里面全是真实的工程任务比如实现一个分布式缓存系统优化一个数据库查询写一个微服务的网关而不是简单的写个排序算法它的PASATD指标就是模型第一次生成的代码直接通过所有测试用力的比例这个指标比PASAT10尝试十次通过更靠谱因为实际开发中你不可能让模型试十次要的是第一次就尽量对Mindforge 27B的PASAT1是49.51%意味着它几乎一半的任务第一次生成的代码就能跑通而Deepseek V4 Pro用了海量代码数据训练参数更大但PASAT1只有47.80%Claude Opus 4.7虽然是顶尖模型但也只比它高1.87个百分点这说明啥说明在编程这种强工程属性的任务里知道怎么开发比知道多少代码片段更重要更有意思的是研究团队做了个对比实验用同样的27B模型一组用1001条完整轨迹训练另一组用100万条代码片段相当于传统数据量训练结果轨迹训练的模型在Program Bench上PASAT1是49.51%而海量片段训练的模型只有32.7%差了快17个百分点这直接证明高质量的全流程数据比海量的局部数据有效得多那这对中小团队意味着啥以前大家觉得要做个能写代码的AI得有千亿参数得有几十万块GPU得收集TB级的数据中小团队根本玩不起但Mindforge 27B告诉我们不用你只需要花精力整理几百上千条高质量的开发轨迹用个小参数模型比如27B几张V100卡就能训练就能做出接近顶尖大模型的效果这对资源有限的团队来说简直是福音不用卷算力不用卷数据量卷数据质量就行了比如一个小团队想做一个金融代码生成的专业模型他们不用去爬Github上所有代码只需要找几个资深金融工程师记录他们开发风控系统交易接口的全过程整理成几百条轨迹就能训练出一个懂金融业务懂工程逻辑的模型这种模型虽然参数小但比那些啥都懂一点的大模型在实际金融场景里更靠谱最后总结一下Mindforge 27B的成功其实打破了一个长期存在的迷思编程能力等于参数规模它证明在编程这种需要强逻辑强工程思维的领域数据的质量是不是全流程是不是包含试错过程是不是有完整上下文比数据量更重要就像教徒弟你让他看一万行代码片段不如让他跟着你完整做十个项目因为前者只能让他记住怎么写后者能让他学会怎么做未来可能会有更多这种小而美的专业模型出现他们不需要千亿参数只需要几千条高质量的领域轨迹就能在特定领域达到顶尖水平这对AI的普及来说是件好事毕竟不是每个团队都有谷歌OpenAI的资源但每个团队都可能拥有高质量的领域数据Mindforge 27B算是给我们开了个好投", "segments": [{"id": 0, "seek": 0, "start": 0.0, "end": 3.18, "text": "Hello 今天咱们来聊点超有意思的AI圈新鲜事", "tokens": [50365, 15947, 220, 12074, 8975, 109, 9497, 6912, 40096, 12579, 19869, 2412, 16697, 1546, 48698, 2523, 230, 12560, 165, 110, 250, 6973, 50524], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 1, "seek": 0, "start": 3.18, "end": 5.8, "text": "绝对能刷新你之前对大模型的刻板印象", "tokens": [50524, 10115, 251, 8713, 8225, 2437, 115, 12560, 2166, 32442, 8713, 3582, 41908, 39823, 1546, 45500, 43664, 35825, 45007, 50655], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 2, "seek": 0, "start": 5.8, "end": 8.26, "text": "今天我们来聊聊一个有点意思的模型", "tokens": [50655, 12074, 15003, 6912, 40096, 40096, 20182, 2412, 12579, 16697, 1546, 41908, 39823, 50778], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 3, "seek": 0, "start": 8.26, "end": 9.8, "text": "Mindforge 27B", "tokens": [50778, 44, 471, 2994, 432, 7634, 33, 50855], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 4, "seek": 0, "start": 9.8, "end": 11.6, "text": "听到名字里的27B", "tokens": [50855, 31022, 4511, 15940, 22381, 15759, 1546, 10076, 33, 50945], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 5, "seek": 0, "start": 11.6, "end": 14.120000000000001, "text": "你可能会觉得这不就是个小参数模型吗", "tokens": [50945, 2166, 16657, 12949, 29992, 5562, 1960, 5620, 7549, 7322, 2129, 224, 33188, 41908, 39823, 14769, 51071], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 6, "seek": 0, "start": 14.120000000000001, "end": 17.38, "text": "毕竟现在动辄千亿参数的AI模型满天飞", "tokens": [51071, 7256, 243, 11957, 253, 25040, 34961, 9830, 226, 20787, 1369, 123, 2129, 224, 33188, 1546, 48698, 41908, 39823, 15868, 94, 6135, 11808, 252, 51234], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 7, "seek": 0, "start": 17.38, "end": 19.34, "text": "27B顶多算个小个子", "tokens": [51234, 10076, 33, 10178, 114, 6392, 19497, 7549, 7322, 7549, 7626, 51332], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 8, "seek": 0, "start": 19.34, "end": 20.86, "text": "但就是这个小个子", "tokens": [51332, 8395, 5620, 15368, 7322, 7549, 7626, 51408], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 9, "seek": 0, "start": 20.86, "end": 23.5, "text": "在编程能力上却干翻了不少大块头", "tokens": [51408, 3581, 38109, 244, 29649, 8225, 13486, 5708, 5322, 112, 26111, 42716, 2289, 1960, 15686, 3582, 47734, 39862, 51540], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 10, "seek": 0, "start": 23.5, "end": 26.02, "text": "在编程测试基准Program Bench上", "tokens": [51540, 3581, 38109, 244, 29649, 11038, 233, 5233, 243, 26008, 6336, 228, 12681, 1342, 3964, 339, 5708, 51666], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 11, "seek": 0, "start": 26.02, "end": 27.76, "text": "它的第一次尝试通过率", "tokens": [51666, 45224, 18049, 9487, 1530, 251, 5233, 243, 19550, 16866, 44866, 51753], "temperature": 0, "avg_logprob": -0.08395666771746696, "compression_ratio": 1.2968299711815563, "no_speech_prob": 2.053063795359744e-11}, {"id": 12, "seek": 2776, "start": 27.76, "end": 30.560000000000002, "text": "Pasit-E达到了49.51%", "tokens": [50365, 47, 296, 270, 12, 36, 9830, 122, 21381, 14938, 13, 18682, 4, 50505], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 13, "seek": 2776, "start": 30.560000000000002, "end": 34.400000000000006, "text": "比DeepSeek第四代专业版的47.80%还高", "tokens": [50505, 11706, 11089, 595, 10637, 916, 7536, 19425, 19105, 940, 241, 940, 248, 42096, 1546, 14060, 13, 4702, 4, 14852, 12979, 50697], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 14, "seek": 2776, "start": 34.400000000000006, "end": 38.86, "text": "甚至逼近了Cloud Opus 4.7的51.38%", "tokens": [50697, 17556, 20844, 2215, 120, 17463, 2289, 32787, 12011, 301, 1017, 13, 22, 1546, 18682, 13, 12625, 4, 50920], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 15, "seek": 2776, "start": 38.86, "end": 42.38, "text": "要知道Cloud Opus 4.7可是公认的编程高手", "tokens": [50920, 4275, 7758, 32787, 12011, 301, 1017, 13, 22, 23359, 13545, 7422, 97, 1546, 38109, 244, 29649, 12979, 11389, 51096], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 16, "seek": 2776, "start": 42.38, "end": 45.94, "text": "而Mindforge 27B只用了1001条数据训练", "tokens": [51096, 11070, 44, 471, 2994, 432, 7634, 33, 14003, 9254, 2289, 6879, 16, 48837, 33188, 26075, 106, 7422, 255, 10115, 225, 51274], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 17, "seek": 2776, "start": 45.94, "end": 47.56, "text": "这到底是咋做到的", "tokens": [51274, 5562, 33883, 1541, 8975, 233, 10907, 4511, 1546, 51355], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 18, "seek": 2776, "start": 47.56, "end": 50.34, "text": "咱们今天就来扒一扒它背后的技术细节", "tokens": [51355, 8975, 109, 9497, 12074, 3111, 6912, 3416, 240, 2257, 3416, 240, 11284, 46329, 13547, 1546, 32502, 1474, 107, 10115, 228, 45161, 51494], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 19, "seek": 2776, "start": 50.34, "end": 51.46, "text": "首先得说", "tokens": [51494, 36490, 5916, 8090, 51550], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 20, "seek": 2776, "start": 51.46, "end": 53.8, "text": "这个模型的核心不在参数多", "tokens": [51550, 15368, 41908, 39823, 1546, 8987, 116, 7945, 1960, 3581, 2129, 224, 33188, 6392, 51667], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 21, "seek": 2776, "start": 53.8, "end": 55.0, "text": "而在数据精", "tokens": [51667, 11070, 3581, 33188, 26075, 106, 34910, 51727], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 22, "seek": 2776, "start": 55.0, "end": 57.540000000000006, "text": "传统的大模型训练编程能力", "tokens": [51727, 7384, 254, 10115, 253, 1546, 3582, 41908, 39823, 7422, 255, 10115, 225, 38109, 244, 29649, 8225, 13486, 51854], "temperature": 0, "avg_logprob": -0.11561867911056434, "compression_ratio": 1.1376146788990826, "no_speech_prob": 1.0420265345034707e-11}, {"id": 23, "seek": 5754, "start": 57.54, "end": 59.32, "text": "基本都是海量堆数据", "tokens": [50365, 37946, 22796, 20078, 26748, 10726, 228, 33188, 26075, 106, 50454], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 24, "seek": 5754, "start": 59.32, "end": 61.22, "text": "GitHub上几百万个项目", "tokens": [50454, 38, 270, 21150, 5708, 6336, 254, 31906, 23570, 7549, 10178, 117, 11386, 50549], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 25, "seek": 5754, "start": 61.22, "end": 62.82, "text": "几十亿行代码片段", "tokens": [50549, 6336, 254, 20145, 1369, 123, 8082, 19105, 23230, 223, 16668, 28427, 50629], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 26, "seek": 5754, "start": 62.82, "end": 65.7, "text": "甚至包括各种开源库的函数定义", "tokens": [50629, 17556, 20844, 41828, 17516, 39810, 18937, 47402, 6346, 241, 1546, 6336, 121, 33188, 12088, 2930, 231, 50773], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 27, "seek": 5754, "start": 65.7, "end": 66.36, "text": "注释", "tokens": [50773, 26432, 5873, 232, 50806], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 28, "seek": 5754, "start": 66.36, "end": 67.4, "text": "用法视力", "tokens": [50806, 9254, 11148, 40656, 13486, 50858], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 29, "seek": 5754, "start": 67.4, "end": 69.0, "text": "一股脑往模型里塞", "tokens": [50858, 2257, 14356, 94, 27067, 239, 29510, 41908, 39823, 15759, 13331, 252, 50938], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 30, "seek": 5754, "start": 69.0, "end": 71.46, "text": "这种训练方式确实能让模型", "tokens": [50938, 5562, 39810, 7422, 255, 10115, 225, 9249, 27584, 38114, 106, 24726, 8225, 33650, 41908, 39823, 51061], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 31, "seek": 5754, "start": 71.46, "end": 72.84, "text": "学会写代码的语法", "tokens": [51061, 29618, 12949, 5676, 247, 19105, 23230, 223, 1546, 5233, 255, 11148, 51130], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 32, "seek": 5754, "start": 72.84, "end": 74.34, "text": "比如怎么定义变量", "tokens": [51130, 36757, 15282, 12088, 2930, 231, 2129, 246, 26748, 51205], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 33, "seek": 5754, "start": 74.34, "end": 75.2, "text": "写循环", "tokens": [51205, 5676, 247, 2172, 103, 8051, 107, 51248], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 34, "seek": 5754, "start": 75.2, "end": 75.84, "text": "掉裤", "tokens": [51248, 29327, 8083, 97, 51280], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 35, "seek": 5754, "start": 75.84, "end": 77.06, "text": "但有个大问题", "tokens": [51280, 8395, 2412, 7549, 3582, 34069, 51341], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 36, "seek": 5754, "start": 77.06, "end": 78.94, "text": "它学到的是局部技能", "tokens": [51341, 11284, 29618, 4511, 24620, 34703, 13470, 32502, 8225, 51435], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 37, "seek": 5754, "start": 78.94, "end": 80.32, "text": "而不是工程思维", "tokens": [51435, 11070, 7296, 23323, 29649, 8870, 10115, 112, 51504], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 38, "seek": 5754, "start": 80.32, "end": 83.0, "text": "就像你让一个只会被单词的人写小说", "tokens": [51504, 3111, 12760, 2166, 33650, 20182, 14003, 12949, 23238, 47446, 5233, 235, 29979, 5676, 247, 7322, 8090, 51638], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 39, "seek": 5754, "start": 83.0, "end": 84.4, "text": "他可能词都认识", "tokens": [51638, 5000, 16657, 5233, 235, 7182, 7422, 97, 5233, 228, 51708], "temperature": 0, "avg_logprob": -0.0922300238358347, "compression_ratio": 1.2413793103448276, "no_speech_prob": 1.4331609683726487e-11}, {"id": 40, "seek": 8440, "start": 84.4, "end": 86.02000000000001, "text": "但写不出连贯的故事", "tokens": [50365, 8395, 5676, 247, 1960, 7781, 3316, 252, 18464, 107, 1546, 43045, 6973, 50446], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 41, "seek": 8440, "start": 86.02000000000001, "end": 88.96000000000001, "text": "Mindforge 27B完全反其道而行之", "tokens": [50446, 44, 471, 2994, 432, 7634, 33, 37100, 22138, 9572, 6025, 11070, 8082, 9574, 50593], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 42, "seek": 8440, "start": 88.96000000000001, "end": 90.88000000000001, "text": "它没用海量代码片段", "tokens": [50593, 11284, 10062, 9254, 20078, 26748, 19105, 23230, 223, 16668, 28427, 50689], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 43, "seek": 8440, "start": 90.88000000000001, "end": 93.28, "text": "而是用了1001条完整开发轨迹", "tokens": [50689, 11070, 1541, 9254, 2289, 6879, 16, 48837, 14128, 27662, 18937, 28926, 17819, 101, 3316, 117, 50809], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 44, "seek": 8440, "start": 93.28, "end": 94.98, "text": "啥叫完整开发轨迹", "tokens": [50809, 3284, 98, 19855, 14128, 27662, 18937, 28926, 17819, 101, 3316, 117, 50894], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 45, "seek": 8440, "start": 94.98, "end": 96.2, "text": "咱们打个比方", "tokens": [50894, 8975, 109, 9497, 12467, 7549, 11706, 9249, 50955], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 46, "seek": 8440, "start": 96.2, "end": 99.4, "text": "你让一个程序员开发一个电商订单管理系统", "tokens": [50955, 2166, 33650, 20182, 29649, 6346, 237, 3606, 246, 18937, 28926, 20182, 42182, 45581, 7422, 95, 47446, 23131, 13876, 25368, 10115, 253, 51115], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 47, "seek": 8440, "start": 99.4, "end": 101.9, "text": "传统数据可能只给他订单创建", "tokens": [51115, 7384, 254, 10115, 253, 33188, 26075, 106, 16657, 14003, 23197, 5000, 7422, 95, 47446, 2437, 249, 34157, 51240], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 48, "seek": 8440, "start": 101.9, "end": 103.36000000000001, "text": "这个函数的代码片段", "tokens": [51240, 15368, 6336, 121, 33188, 1546, 19105, 23230, 223, 16668, 28427, 51313], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 49, "seek": 8440, "start": 103.36000000000001, "end": 105.44, "text": "或者库存检查的一段逻辑", "tokens": [51313, 31148, 6346, 241, 39781, 16407, 222, 42623, 1546, 2257, 28427, 2215, 119, 9830, 239, 51417], "temperature": 0, "avg_logprob": -0.07542677302109568, "compression_ratio": 1.2074074074074075, "no_speech_prob": 1.5084575949453338e-11}, {"id": 50, "seek": 10544, "start": 105.44, "end": 107.1, "text": "但完整开发轨迹", "tokens": [50365, 8395, 14128, 27662, 18937, 28926, 17819, 101, 3316, 117, 50448], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 51, "seek": 10544, "start": 107.1, "end": 108.89999999999999, "text": "是从老板提需求开始", "tokens": [50448, 1541, 35630, 10439, 43664, 20949, 32535, 32718, 45213, 50538], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 52, "seek": 10544, "start": 108.89999999999999, "end": 110.67999999999999, "text": "到程序员分析需求", "tokens": [50538, 4511, 29649, 6346, 237, 3606, 246, 6627, 7360, 238, 32535, 32718, 50627], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 53, "seek": 10544, "start": 110.67999999999999, "end": 111.92, "text": "设计系统架构", "tokens": [50627, 7422, 122, 7422, 94, 25368, 10115, 253, 7360, 114, 7360, 226, 50689], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 54, "seek": 10544, "start": 111.92, "end": 112.94, "text": "画流程图", "tokens": [50689, 27126, 27854, 29649, 3919, 122, 50740], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 55, "seek": 10544, "start": 112.94, "end": 113.72, "text": "写代码", "tokens": [50740, 5676, 247, 19105, 23230, 223, 50779], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 56, "seek": 10544, "start": 113.72, "end": 114.36, "text": "测bug", "tokens": [50779, 11038, 233, 44455, 50811], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 57, "seek": 10544, "start": 114.36, "end": 115.03999999999999, "text": "改bug", "tokens": [50811, 34490, 44455, 50845], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 58, "seek": 10544, "start": 115.03999999999999, "end": 115.98, "text": "最后上线", "tokens": [50845, 8661, 13547, 5708, 16853, 123, 50892], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 59, "seek": 10544, "start": 115.98, "end": 118.0, "text": "整个过程每一步的对话", "tokens": [50892, 27662, 7549, 16866, 29649, 23664, 2257, 31429, 1546, 8713, 21596, 50993], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 60, "seek": 10544, "start": 118.0, "end": 118.62, "text": "思考", "tokens": [50993, 8870, 26504, 51024], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 61, "seek": 10544, "start": 118.62, "end": 119.52, "text": "代码修改", "tokens": [51024, 19105, 23230, 223, 7792, 106, 34490, 51069], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 62, "seek": 10544, "start": 119.52, "end": 120.36, "text": "错误反馈", "tokens": [51069, 29900, 5233, 107, 22138, 11748, 230, 51111], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 63, "seek": 10544, "start": 120.36, "end": 121.42, "text": "全都有记录", "tokens": [51111, 11319, 48121, 34756, 7391, 243, 51164], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 64, "seek": 10544, "start": 121.42, "end": 122.56, "text": "具体来说", "tokens": [51164, 39806, 29485, 6912, 8090, 51221], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 65, "seek": 10544, "start": 122.56, "end": 124.0, "text": "这1001条轨迹", "tokens": [51221, 5562, 6879, 16, 48837, 17819, 101, 3316, 117, 51293], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 66, "seek": 10544, "start": 124.0, "end": 126.74, "text": "平均每条有181.6轮对话", "tokens": [51293, 16716, 14872, 229, 23664, 48837, 2412, 6494, 16, 13, 21, 17819, 106, 8713, 21596, 51430], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 67, "seek": 10544, "start": 126.74, "end": 128.22, "text": "这不是简单的问答", "tokens": [51430, 5562, 7296, 11249, 222, 47446, 1546, 22064, 28223, 51504], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 68, "seek": 10544, "start": 128.22, "end": 130.56, "text": "而是像真实开发团队里的协作", "tokens": [51504, 11070, 1541, 12760, 6303, 24726, 18937, 28926, 3919, 95, 10034, 253, 15759, 1546, 5322, 237, 11914, 51621], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 69, "seek": 10544, "start": 130.56, "end": 131.68, "text": "比如第一轮", "tokens": [51621, 36757, 18049, 17819, 106, 51677], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 70, "seek": 10544, "start": 131.68, "end": 133.38, "text": "用户也就是需求方说", "tokens": [51677, 9254, 1486, 115, 6404, 5620, 32535, 32718, 9249, 8090, 51762], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 71, "seek": 10544, "start": 133.38, "end": 134.98, "text": "我们需要一个订单系统", "tokens": [51762, 15003, 35748, 20182, 7422, 95, 47446, 25368, 10115, 253, 51842], "temperature": 0, "avg_logprob": -0.08794936393071147, "compression_ratio": 1.276190476190476, "no_speech_prob": 1.600414592517474e-11}, {"id": 72, "seek": 13498, "start": 134.98, "end": 136.79999999999998, "text": "要支持多种支付方式", "tokens": [50365, 4275, 35488, 6392, 39810, 24400, 41469, 9249, 27584, 50456], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 73, "seek": 13498, "start": 136.79999999999998, "end": 138.45999999999998, "text": "还要能实时查库存", "tokens": [50456, 14852, 4275, 8225, 24726, 15729, 42623, 6346, 241, 39781, 50539], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 74, "seek": 13498, "start": 138.45999999999998, "end": 140.01999999999998, "text": "模型作为开发者", "tokens": [50539, 41908, 39823, 11914, 13992, 18937, 28926, 12444, 50617], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 75, "seek": 13498, "start": 140.01999999999998, "end": 141.35999999999999, "text": "不会直接写代码", "tokens": [50617, 47928, 43297, 5676, 247, 19105, 23230, 223, 50684], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 76, "seek": 13498, "start": 141.35999999999999, "end": 142.38, "text": "而是先问细节", "tokens": [50684, 11070, 1541, 10108, 22064, 10115, 228, 45161, 50735], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 77, "seek": 13498, "start": 142.38, "end": 144.66, "text": "支付方式需要支持支付宝", "tokens": [50735, 24400, 41469, 9249, 27584, 35748, 35488, 24400, 41469, 2415, 251, 50849], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 78, "seek": 13498, "start": 144.66, "end": 145.98, "text": "微信还是银行卡", "tokens": [50849, 39152, 17665, 45726, 165, 241, 114, 8082, 32681, 50915], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 79, "seek": 13498, "start": 145.98, "end": 148.33999999999997, "text": "实时查库存是扣库存钱查", "tokens": [50915, 24726, 15729, 42623, 6346, 241, 39781, 1541, 3416, 96, 6346, 241, 39781, 39623, 42623, 51033], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 80, "seek": 13498, "start": 148.33999999999997, "end": 149.56, "text": "还是扣库存实查", "tokens": [51033, 45726, 3416, 96, 6346, 241, 39781, 24726, 42623, 51094], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 81, "seek": 13498, "start": 149.56, "end": 150.72, "text": "订单取消后", "tokens": [51094, 7422, 95, 47446, 29436, 28837, 13547, 51152], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 82, "seek": 13498, "start": 150.72, "end": 151.79999999999998, "text": "库存要不要回退", "tokens": [51152, 6346, 241, 39781, 4275, 11962, 8350, 46361, 51206], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 83, "seek": 13498, "start": 151.79999999999998, "end": 153.01999999999998, "text": "用户回答后", "tokens": [51206, 9254, 1486, 115, 8350, 28223, 13547, 51267], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 84, "seek": 13498, "start": 153.01999999999998, "end": 154.6, "text": "模型会接着设计架构", "tokens": [51267, 41908, 39823, 12949, 14468, 20708, 7422, 122, 7422, 94, 7360, 114, 7360, 226, 51346], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 85, "seek": 13498, "start": 154.6, "end": 156.28, "text": "我们可以分订单服务", "tokens": [51346, 15003, 6723, 6627, 7422, 95, 47446, 27408, 5087, 94, 51430], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 86, "seek": 13498, "start": 156.28, "end": 157.16, "text": "支付服务", "tokens": [51430, 24400, 41469, 27408, 5087, 94, 51474], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 87, "seek": 13498, "start": 157.16, "end": 158.6, "text": "库存服务三个模块", "tokens": [51474, 6346, 241, 39781, 27408, 5087, 94, 10960, 7549, 41908, 47734, 51546], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 88, "seek": 13498, "start": 158.6, "end": 160.16, "text": "用消息对列结偶", "tokens": [51546, 9254, 28837, 26460, 8713, 43338, 45641, 7437, 114, 51624], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 89, "seek": 13498, "start": 160.16, "end": 161.51999999999998, "text": "然后写代码的时候", "tokens": [51624, 26636, 5676, 247, 19105, 23230, 223, 49873, 51692], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 90, "seek": 13498, "start": 161.51999999999998, "end": 163.82, "text": "模型会写订单表的创建语句", "tokens": [51692, 41908, 39823, 12949, 5676, 247, 7422, 95, 47446, 17571, 1546, 2437, 249, 34157, 5233, 255, 34592, 51807], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 91, "seek": 13498, "start": 163.82, "end": 164.95999999999998, "text": "写接口定义", "tokens": [51807, 5676, 247, 14468, 18144, 12088, 2930, 231, 51864], "temperature": 0, "avg_logprob": -0.09175969061450423, "compression_ratio": 1.5250836120401339, "no_speech_prob": 1.4393833347448037e-11}, {"id": 92, "seek": 16496, "start": 164.96, "end": 166.0, "text": "写业务逻辑", "tokens": [50365, 5676, 247, 940, 248, 5087, 94, 2215, 119, 9830, 239, 50417], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 93, "seek": 16496, "start": 166.0, "end": 167.18, "text": "写完跑测试", "tokens": [50417, 5676, 247, 14128, 32585, 11038, 233, 5233, 243, 50476], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 94, "seek": 16496, "start": 167.18, "end": 168.8, "text": "比如发现高并发时", "tokens": [50476, 36757, 28926, 20204, 12979, 3509, 114, 28926, 15729, 50557], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 95, "seek": 16496, "start": 168.8, "end": 169.82000000000002, "text": "库存扣减不对", "tokens": [50557, 6346, 241, 39781, 3416, 96, 6336, 237, 41639, 50608], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 96, "seek": 16496, "start": 169.82000000000002, "end": 172.3, "text": "模型会根据错误日志分析原因", "tokens": [50608, 41908, 39823, 12949, 31337, 26075, 106, 29900, 5233, 107, 6890, 44700, 6627, 7360, 238, 42577, 50732], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 97, "seek": 16496, "start": 172.3, "end": 174.0, "text": "应该是没加分布是锁", "tokens": [50732, 44297, 44646, 1541, 10062, 9990, 6627, 34688, 1541, 23049, 223, 50817], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 98, "seek": 16496, "start": 174.0, "end": 175.14000000000001, "text": "导致两个请求", "tokens": [50817, 4510, 120, 6784, 112, 36257, 7549, 27908, 32718, 50874], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 99, "seek": 16496, "start": 175.14000000000001, "end": 176.60000000000002, "text": "同时读到库存为一", "tokens": [50874, 13089, 15729, 5233, 119, 4511, 6346, 241, 39781, 13992, 2257, 50947], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 100, "seek": 16496, "start": 176.60000000000002, "end": 177.44, "text": "都扣了", "tokens": [50947, 7182, 3416, 96, 2289, 50989], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 101, "seek": 16496, "start": 177.44, "end": 178.54000000000002, "text": "然后修改代码", "tokens": [50989, 26636, 7792, 106, 34490, 19105, 23230, 223, 51044], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 102, "seek": 16496, "start": 178.54000000000002, "end": 179.16, "text": "加锁", "tokens": [51044, 9990, 23049, 223, 51075], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 103, "seek": 16496, "start": 179.16, "end": 179.98000000000002, "text": "再测试", "tokens": [51075, 8623, 11038, 233, 5233, 243, 51116], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 104, "seek": 16496, "start": 179.98000000000002, "end": 180.92000000000002, "text": "直到通过", "tokens": [51116, 16186, 4511, 19550, 16866, 51163], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 105, "seek": 16496, "start": 180.92000000000002, "end": 181.46, "text": "你看", "tokens": [51163, 16529, 51190], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 106, "seek": 16496, "start": 181.46, "end": 183.66, "text": "这181.6轮对话里", "tokens": [51190, 5562, 6494, 16, 13, 21, 17819, 106, 8713, 21596, 15759, 51300], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 107, "seek": 16496, "start": 183.66, "end": 185.72, "text": "每一轮都包含三类信息", "tokens": [51300, 23664, 2257, 17819, 106, 7182, 23305, 2392, 104, 10960, 22113, 119, 17665, 26460, 51403], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 108, "seek": 16496, "start": 185.72, "end": 186.54000000000002, "text": "上下文", "tokens": [51403, 5708, 4438, 17174, 51444], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 109, "seek": 16496, "start": 186.54000000000002, "end": 187.92000000000002, "text": "之前的需求讨论", "tokens": [51444, 32442, 1546, 32535, 32718, 7422, 101, 7422, 118, 51513], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 110, "seek": 16496, "start": 187.92000000000002, "end": 188.86, "text": "架构设计", "tokens": [51513, 7360, 114, 7360, 226, 7422, 122, 7422, 94, 51560], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 111, "seek": 16496, "start": 188.86, "end": 189.56, "text": "动作", "tokens": [51560, 34961, 11914, 51595], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 112, "seek": 16496, "start": 189.56, "end": 190.58, "text": "写了什么代码", "tokens": [51595, 5676, 247, 2289, 10440, 19105, 23230, 223, 51646], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 113, "seek": 16496, "start": 190.58, "end": 191.36, "text": "改了哪里", "tokens": [51646, 34490, 2289, 17028, 15759, 51685], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 114, "seek": 16496, "start": 191.36, "end": 191.98000000000002, "text": "反馈", "tokens": [51685, 22138, 11748, 230, 51716], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 115, "seek": 16496, "start": 191.98000000000002, "end": 193.22, "text": "测试报了什么错", "tokens": [51716, 11038, 233, 5233, 243, 49817, 2289, 10440, 29900, 51778], "temperature": 0, "avg_logprob": -0.09331738253879975, "compression_ratio": 1.290625, "no_speech_prob": 1.6233002789189932e-11}, {"id": 116, "seek": 19322, "start": 193.22, "end": 194.58, "text": "用户提了什么意见", "tokens": [50365, 9254, 1486, 115, 20949, 2289, 10440, 9042, 23813, 50433], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 117, "seek": 19322, "start": 194.58, "end": 195.9, "text": "模型训练的时候", "tokens": [50433, 41908, 39823, 7422, 255, 10115, 225, 49873, 50499], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 118, "seek": 19322, "start": 195.9, "end": 198.28, "text": "就是学习这三者之间的因果链", "tokens": [50499, 5620, 29618, 2930, 254, 5562, 10960, 12444, 9574, 31685, 1546, 8698, 9319, 165, 241, 122, 50618], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 119, "seek": 19322, "start": 198.28, "end": 199.42, "text": "根据上下文", "tokens": [50618, 31337, 26075, 106, 5708, 4438, 17174, 50675], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 120, "seek": 19322, "start": 199.42, "end": 200.6, "text": "应该做什么动作", "tokens": [50675, 44297, 44646, 10907, 10440, 34961, 11914, 50734], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 121, "seek": 19322, "start": 200.6, "end": 202.57999999999998, "text": "做了动作会得到什么反馈", "tokens": [50734, 10907, 2289, 34961, 11914, 12949, 5916, 4511, 10440, 22138, 11748, 230, 50833], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 122, "seek": 19322, "start": 202.57999999999998, "end": 204.54, "text": "得到反馈后该怎么调整", "tokens": [50833, 5916, 4511, 22138, 11748, 230, 13547, 44646, 15282, 8897, 225, 27662, 50931], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 123, "seek": 19322, "start": 204.54, "end": 206.42, "text": "这比单纯学代码片段", "tokens": [50931, 5562, 11706, 47446, 16853, 107, 29618, 19105, 23230, 223, 16668, 28427, 51025], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 124, "seek": 19322, "start": 206.42, "end": 207.12, "text": "长什么样", "tokens": [51025, 32271, 10440, 14496, 51060], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 125, "seek": 19322, "start": 207.12, "end": 208.07999999999998, "text": "高明太多了", "tokens": [51060, 12979, 11100, 9455, 6392, 2289, 51108], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 126, "seek": 19322, "start": 208.07999999999998, "end": 209.12, "text": "他学到的是", "tokens": [51108, 5000, 29618, 4511, 24620, 51160], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 127, "seek": 19322, "start": 209.12, "end": 211.06, "text": "开发代码的全流程逻辑", "tokens": [51160, 18937, 28926, 19105, 23230, 223, 1546, 11319, 27854, 29649, 2215, 119, 9830, 239, 51257], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 128, "seek": 19322, "start": 211.06, "end": 212.52, "text": "也就是工程思维", "tokens": [51257, 6404, 5620, 23323, 29649, 8870, 10115, 112, 51330], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 129, "seek": 19322, "start": 212.52, "end": 214.74, "text": "那这些轨迹数据是怎么来的", "tokens": [51330, 4184, 5562, 13824, 17819, 101, 3316, 117, 33188, 26075, 106, 1541, 15282, 6912, 1546, 51441], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 130, "seek": 19322, "start": 214.74, "end": 216.64, "text": "研究团队从开源项目的", "tokens": [51441, 23230, 242, 44704, 3919, 95, 10034, 253, 35630, 18937, 47402, 10178, 117, 11386, 1546, 51536], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 131, "seek": 19322, "start": 216.64, "end": 218.44, "text": "完整开发历史里挖出来的", "tokens": [51536, 14128, 27662, 18937, 28926, 5014, 228, 45399, 15759, 8501, 244, 44561, 1546, 51626], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 132, "seek": 19322, "start": 218.44, "end": 220.6, "text": "比如Github上一些知名项目的", "tokens": [51626, 36757, 38, 355, 836, 5708, 38515, 6498, 15940, 10178, 117, 11386, 1546, 51734], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 133, "seek": 19322, "start": 220.6, "end": 221.57999999999998, "text": "issue讨论区", "tokens": [51734, 891, 622, 7422, 101, 7422, 118, 9937, 118, 51783], "temperature": 0, "avg_logprob": -0.08094231210103849, "compression_ratio": 1.3492537313432835, "no_speech_prob": 1.4862867880882646e-11}, {"id": 134, "seek": 22158, "start": 221.58, "end": 223.3, "text": "PolRequest的修改记录", "tokens": [50365, 47, 401, 8524, 20343, 1546, 7792, 106, 34490, 34756, 7391, 243, 50451], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 135, "seek": 22158, "start": 223.3, "end": 224.66000000000003, "text": "代码评审的评论", "tokens": [50451, 19105, 23230, 223, 5233, 226, 2415, 94, 1546, 5233, 226, 7422, 118, 50519], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 136, "seek": 22158, "start": 224.66000000000003, "end": 227.0, "text": "甚至是开发者之间的聊天记录", "tokens": [50519, 17556, 20844, 1541, 18937, 28926, 12444, 9574, 31685, 1546, 40096, 6135, 34756, 7391, 243, 50636], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 137, "seek": 22158, "start": 227.0, "end": 228.0, "text": "当然脱敏了", "tokens": [50636, 40486, 27067, 109, 7017, 237, 2289, 50686], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 138, "seek": 22158, "start": 228.0, "end": 230.06, "text": "他们把这些非结构化的数据", "tokens": [50686, 47911, 16075, 5562, 13824, 12107, 45641, 7360, 226, 23756, 1546, 33188, 26075, 106, 50789], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 139, "seek": 22158, "start": 230.06, "end": 231.76000000000002, "text": "整理成结构化的对话", "tokens": [50789, 27662, 13876, 11336, 45641, 7360, 226, 23756, 1546, 8713, 21596, 50874], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 140, "seek": 22158, "start": 231.76000000000002, "end": 232.36, "text": "代码", "tokens": [50874, 19105, 23230, 223, 50904], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 141, "seek": 22158, "start": 232.36, "end": 233.16000000000003, "text": "反馈链", "tokens": [50904, 22138, 11748, 230, 165, 241, 122, 50944], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 142, "seek": 22158, "start": 233.16000000000003, "end": 235.38000000000002, "text": "每条轨迹都像一部开发日志", "tokens": [50944, 23664, 48837, 17819, 101, 3316, 117, 7182, 12760, 2257, 13470, 18937, 28926, 6890, 44700, 51055], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 143, "seek": 22158, "start": 235.38000000000002, "end": 236.62, "text": "记录了一个功能", "tokens": [51055, 34756, 7391, 243, 2289, 20182, 32311, 8225, 51117], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 144, "seek": 22158, "start": 236.62, "end": 238.10000000000002, "text": "从无到有的全过程", "tokens": [51117, 35630, 31364, 4511, 2412, 1546, 11319, 16866, 29649, 51191], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 145, "seek": 22158, "start": 238.10000000000002, "end": 239.24, "text": "比如某个轨迹", "tokens": [51191, 36757, 17238, 238, 7549, 17819, 101, 3316, 117, 51248], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 146, "seek": 22158, "start": 239.24, "end": 240.3, "text": "可能是开发一个", "tokens": [51248, 16657, 1541, 18937, 28926, 20182, 51301], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 147, "seek": 22158, "start": 240.3, "end": 241.88000000000002, "text": "用户登陆健全功能", "tokens": [51301, 9254, 1486, 115, 25874, 8842, 228, 44201, 11319, 32311, 8225, 51380], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 148, "seek": 22158, "start": 241.88000000000002, "end": 243.12, "text": "从最开始讨论", "tokens": [51380, 35630, 8661, 45213, 7422, 101, 7422, 118, 51442], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 149, "seek": 22158, "start": 243.12, "end": 244.9, "text": "用JWT还是Session", "tokens": [51442, 9254, 41, 54, 51, 45726, 50, 4311, 51531], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 150, "seek": 22158, "start": 244.9, "end": 246.82000000000002, "text": "到设计Token刷新机制", "tokens": [51531, 4511, 7422, 122, 7422, 94, 51, 8406, 2437, 115, 12560, 37960, 25491, 51627], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 151, "seek": 22158, "start": 246.82000000000002, "end": 248.48000000000002, "text": "到写登陆接口代码", "tokens": [51627, 4511, 5676, 247, 25874, 8842, 228, 14468, 18144, 19105, 23230, 223, 51710], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 152, "seek": 22158, "start": 248.48000000000002, "end": 249.58, "text": "到测试发现", "tokens": [51710, 4511, 11038, 233, 5233, 243, 28926, 20204, 51765], "temperature": 0, "avg_logprob": -0.0837698841546949, "compression_ratio": 1.3046153846153845, "no_speech_prob": 1.878600400240238e-11}, {"id": 153, "seek": 24958, "start": 249.58, "end": 251.12, "text": "Token过期时间太短", "tokens": [50365, 51, 8406, 16866, 16786, 44848, 9455, 5881, 255, 50442], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 154, "seek": 24958, "start": 251.12, "end": 252.74, "text": "再到调整过期时间", "tokens": [50442, 8623, 4511, 8897, 225, 27662, 16866, 16786, 44848, 50523], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 155, "seek": 24958, "start": 252.74, "end": 253.98000000000002, "text": "加刷新接口", "tokens": [50523, 9990, 2437, 115, 12560, 14468, 18144, 50585], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 156, "seek": 24958, "start": 253.98000000000002, "end": 254.86, "text": "最后上线", "tokens": [50585, 8661, 13547, 5708, 16853, 123, 50629], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 157, "seek": 24958, "start": 254.86, "end": 255.86, "text": "这些细节", "tokens": [50629, 5562, 13824, 10115, 228, 45161, 50679], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 158, "seek": 24958, "start": 255.86, "end": 257.24, "text": "传统代码数据里", "tokens": [50679, 7384, 254, 10115, 253, 19105, 23230, 223, 33188, 26075, 106, 15759, 50748], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 159, "seek": 24958, "start": 257.24, "end": 258.24, "text": "根本不会保留", "tokens": [50748, 31337, 8802, 47928, 24302, 24456, 50798], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 160, "seek": 24958, "start": 258.24, "end": 259.86, "text": "接下来是训练策略", "tokens": [50798, 14468, 4438, 6912, 1541, 7422, 255, 10115, 225, 7973, 244, 6904, 98, 50879], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 161, "seek": 24958, "start": 259.86, "end": 261.1, "text": "研究团队用了", "tokens": [50879, 23230, 242, 44704, 3919, 95, 10034, 253, 9254, 2289, 50941], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 162, "seek": 24958, "start": 261.1, "end": 262.2, "text": "英国语言建模", "tokens": [50941, 27869, 16086, 5233, 255, 12009, 34157, 41908, 50996], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 163, "seek": 24958, "start": 262.2, "end": 264.18, "text": "但不是普通的英国建模", "tokens": [50996, 8395, 7296, 29993, 19550, 1546, 27869, 16086, 34157, 41908, 51095], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 164, "seek": 24958, "start": 264.18, "end": 265.3, "text": "普通建模是", "tokens": [51095, 29993, 19550, 34157, 41908, 1541, 51151], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 165, "seek": 24958, "start": 265.3, "end": 266.94, "text": "把一段文本拆成Token", "tokens": [51151, 16075, 2257, 28427, 17174, 8802, 6852, 228, 11336, 51, 8406, 51233], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 166, "seek": 24958, "start": 266.94, "end": 268.90000000000003, "text": "让模型预测下一个Token", "tokens": [51233, 33650, 41908, 39823, 12501, 226, 11038, 233, 4438, 20182, 51, 8406, 51331], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 167, "seek": 24958, "start": 268.90000000000003, "end": 270.68, "text": "而MindForge 27B", "tokens": [51331, 11070, 44, 471, 37, 4685, 7634, 33, 51420], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 168, "seek": 24958, "start": 270.68, "end": 272.16, "text": "把轨迹拆成片段", "tokens": [51420, 16075, 17819, 101, 3316, 117, 6852, 228, 11336, 16668, 28427, 51494], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 169, "seek": 24958, "start": 272.16, "end": 273.04, "text": "每个片段", "tokens": [51494, 23664, 7549, 16668, 28427, 51538], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 170, "seek": 24958, "start": 273.04, "end": 274.42, "text": "可能是用户需求", "tokens": [51538, 16657, 1541, 9254, 1486, 115, 32535, 32718, 51607], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 171, "seek": 24958, "start": 274.42, "end": 275.38, "text": "模型回复", "tokens": [51607, 41908, 39823, 8350, 1787, 235, 51655], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 172, "seek": 24958, "start": 275.38, "end": 276.14, "text": "代码快", "tokens": [51655, 19105, 23230, 223, 10251, 51693], "temperature": 0, "avg_logprob": -0.08976845992238898, "compression_ratio": 1.2450331125827814, "no_speech_prob": 1.9608425994022127e-11}, {"id": 173, "seek": 27614, "start": 276.14, "end": 277.0, "text": "错误日志", "tokens": [50365, 29900, 5233, 107, 6890, 44700, 50408], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 174, "seek": 27614, "start": 277.0, "end": 279.78, "text": "然后让模型学习在什么上下文下", "tokens": [50408, 26636, 33650, 41908, 39823, 29618, 2930, 254, 3581, 10440, 5708, 4438, 17174, 4438, 50547], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 175, "seek": 27614, "start": 279.78, "end": 281.59999999999997, "text": "该生成什么类型的片段", "tokens": [50547, 44646, 8244, 11336, 10440, 22113, 119, 39823, 1546, 16668, 28427, 50638], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 176, "seek": 27614, "start": 281.59999999999997, "end": 284.21999999999997, "text": "比如前面是用户提了一个bug反馈", "tokens": [50638, 36757, 8945, 8833, 1541, 9254, 1486, 115, 20949, 2289, 20182, 44455, 22138, 11748, 230, 50769], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 177, "seek": 27614, "start": 284.21999999999997, "end": 287.56, "text": "模型就该预测错误分析和代码修改片段", "tokens": [50769, 41908, 39823, 3111, 44646, 12501, 226, 11038, 233, 29900, 5233, 107, 6627, 7360, 238, 12565, 19105, 23230, 223, 7792, 106, 34490, 16668, 28427, 50936], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 178, "seek": 27614, "start": 287.56, "end": 289.7, "text": "而不是直接生成代码片段", "tokens": [50936, 11070, 7296, 43297, 8244, 11336, 19105, 23230, 223, 16668, 28427, 51043], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 179, "seek": 27614, "start": 289.7, "end": 293.5, "text": "这种结构化预测让模型更理解开发流程的节奏", "tokens": [51043, 5562, 39810, 45641, 7360, 226, 23756, 12501, 226, 11038, 233, 33650, 41908, 39823, 19002, 13876, 17278, 18937, 28926, 27854, 29649, 1546, 45161, 1881, 237, 51233], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 180, "seek": 27614, "start": 293.5, "end": 294.82, "text": "还有一个关键点", "tokens": [51233, 35091, 20182, 28053, 23049, 106, 12579, 51299], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 181, "seek": 27614, "start": 294.82, "end": 297.14, "text": "模型在训练时会模拟错误", "tokens": [51299, 41908, 39823, 3581, 7422, 255, 10115, 225, 15729, 12949, 41908, 6852, 253, 29900, 5233, 107, 51415], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 182, "seek": 27614, "start": 297.14, "end": 298.26, "text": "传统训练里", "tokens": [51415, 7384, 254, 10115, 253, 7422, 255, 10115, 225, 15759, 51471], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 183, "seek": 27614, "start": 298.26, "end": 300.12, "text": "代码数据大多是正确的", "tokens": [51471, 19105, 23230, 223, 33188, 26075, 106, 3582, 6392, 1541, 15789, 38114, 106, 1546, 51564], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 184, "seek": 27614, "start": 300.12, "end": 301.97999999999996, "text": "比如开源项目的最终代码", "tokens": [51564, 36757, 18937, 47402, 10178, 117, 11386, 1546, 8661, 10115, 230, 19105, 23230, 223, 51657], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 185, "seek": 27614, "start": 301.97999999999996, "end": 304.76, "text": "模型很少看到错误代码和修复过程", "tokens": [51657, 41908, 39823, 4563, 15686, 18032, 29900, 5233, 107, 19105, 23230, 223, 12565, 7792, 106, 1787, 235, 16866, 29649, 51796], "temperature": 0, "avg_logprob": -0.07240762758017773, "compression_ratio": 1.428115015974441, "no_speech_prob": 1.461946015635096e-11}, {"id": 186, "seek": 30476, "start": 304.76, "end": 307.08, "text": "但MindForge 27B的轨迹里", "tokens": [50365, 8395, 44, 471, 37, 4685, 7634, 33, 1546, 17819, 101, 3316, 117, 15759, 50481], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 187, "seek": 30476, "start": 307.08, "end": 309.64, "text": "包含大量错误修复的循环", "tokens": [50481, 23305, 2392, 104, 3582, 26748, 29900, 5233, 107, 7792, 106, 1787, 235, 1546, 2172, 103, 8051, 107, 50609], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 188, "seek": 30476, "start": 309.64, "end": 311.26, "text": "比如模型写了一段代码", "tokens": [50609, 36757, 41908, 39823, 5676, 247, 2289, 2257, 28427, 19105, 23230, 223, 50690], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 189, "seek": 30476, "start": 311.26, "end": 312.94, "text": "测试爆空指针异常", "tokens": [50690, 11038, 233, 5233, 243, 45827, 23322, 25922, 25153, 230, 5702, 224, 11279, 50774], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 190, "seek": 30476, "start": 312.94, "end": 315.62, "text": "然后模型会分析哪个变量可能为空", "tokens": [50774, 26636, 41908, 39823, 12949, 6627, 7360, 238, 17028, 7549, 2129, 246, 26748, 16657, 13992, 23322, 50908], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 191, "seek": 30476, "start": 315.62, "end": 317.21999999999997, "text": "修改代码加判空", "tokens": [50908, 7792, 106, 34490, 19105, 23230, 223, 9990, 2437, 97, 23322, 50988], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 192, "seek": 30476, "start": 317.21999999999997, "end": 318.08, "text": "再测试", "tokens": [50988, 8623, 11038, 233, 5233, 243, 51031], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 193, "seek": 30476, "start": 318.08, "end": 319.74, "text": "可能又爆类型不匹配", "tokens": [51031, 16657, 17047, 45827, 22113, 119, 39823, 1960, 9937, 117, 38846, 51114], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 194, "seek": 30476, "start": 319.74, "end": 320.7, "text": "再调整", "tokens": [51114, 8623, 8897, 225, 27662, 51162], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 195, "seek": 30476, "start": 320.7, "end": 321.78, "text": "这种试错过程", "tokens": [51162, 5562, 39810, 5233, 243, 29900, 16866, 29649, 51216], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 196, "seek": 30476, "start": 321.78, "end": 323.65999999999997, "text": "让模型学会了如何调试", "tokens": [51216, 33650, 41908, 39823, 29618, 12949, 2289, 43526, 8897, 225, 5233, 243, 51310], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 197, "seek": 30476, "start": 323.65999999999997, "end": 325.46, "text": "这不是靠记住常见bug", "tokens": [51310, 5562, 7296, 5363, 254, 34756, 21632, 11279, 23813, 44455, 51400], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 198, "seek": 30476, "start": 325.46, "end": 327.71999999999997, "text": "而是掌握了排查问题的思路", "tokens": [51400, 11070, 1541, 6900, 234, 11673, 94, 2289, 44647, 42623, 34069, 1546, 8870, 24658, 51513], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 199, "seek": 30476, "start": 327.71999999999997, "end": 329.15999999999997, "text": "就像一个老程序员", "tokens": [51513, 3111, 12760, 20182, 10439, 29649, 6346, 237, 3606, 246, 51585], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 200, "seek": 30476, "start": 329.15999999999997, "end": 331.18, "text": "不是背会了所有bug的解法", "tokens": [51585, 7296, 46329, 12949, 2289, 39300, 44455, 1546, 17278, 11148, 51686], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 201, "seek": 30476, "start": 331.18, "end": 333.34, "text": "而是掌握了排查问题的思路", "tokens": [51686, 11070, 1541, 6900, 234, 11673, 94, 2289, 44647, 42623, 34069, 1546, 8870, 24658, 51794], "temperature": 0, "avg_logprob": -0.08423756136752591, "compression_ratio": 1.3395638629283488, "no_speech_prob": 1.8017750486043482e-11}, {"id": 202, "seek": 33334, "start": 333.34, "end": 334.78, "text": "那效果为啥这么好", "tokens": [50365, 4184, 43076, 9319, 13992, 3284, 98, 31889, 2131, 50437], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 203, "seek": 33334, "start": 334.78, "end": 336.23999999999995, "text": "咱们再看实验细节", "tokens": [50437, 8975, 109, 9497, 8623, 4200, 24726, 49657, 234, 10115, 228, 45161, 50510], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 204, "seek": 33334, "start": 336.23999999999995, "end": 339.34, "text": "ProgramBench是一个很严格的编程测试集", "tokens": [50510, 12681, 1342, 21736, 339, 1541, 20182, 4563, 940, 98, 30921, 1546, 38109, 244, 29649, 11038, 233, 5233, 243, 26020, 50665], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 205, "seek": 33334, "start": 339.34, "end": 341.52, "text": "里面全是真实的工程任务", "tokens": [50665, 15759, 8833, 11319, 1541, 6303, 24726, 1546, 23323, 29649, 26443, 5087, 94, 50774], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 206, "seek": 33334, "start": 341.52, "end": 344.06, "text": "比如实现一个分布式缓存系统", "tokens": [50774, 36757, 24726, 20204, 20182, 6627, 34688, 27584, 38109, 241, 39781, 25368, 10115, 253, 50901], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 207, "seek": 33334, "start": 344.06, "end": 345.82, "text": "优化一个数据库查询", "tokens": [50901, 7384, 246, 23756, 20182, 33188, 26075, 106, 6346, 241, 42623, 5233, 95, 50989], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 208, "seek": 33334, "start": 345.82, "end": 347.5, "text": "写一个微服务的网关", "tokens": [50989, 5676, 247, 20182, 39152, 27408, 5087, 94, 1546, 16469, 239, 28053, 51073], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 209, "seek": 33334, "start": 347.5, "end": 349.76, "text": "而不是简单的写个排序算法", "tokens": [51073, 11070, 7296, 11249, 222, 47446, 1546, 5676, 247, 7549, 44647, 6346, 237, 19497, 11148, 51186], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 210, "seek": 33334, "start": 349.76, "end": 351.29999999999995, "text": "它的PASATD指标", "tokens": [51186, 45224, 47, 3160, 2218, 35, 25922, 162, 3921, 51263], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 211, "seek": 33334, "start": 351.29999999999995, "end": 353.55999999999995, "text": "就是模型第一次生成的代码", "tokens": [51263, 5620, 41908, 39823, 18049, 9487, 8244, 11336, 1546, 19105, 23230, 223, 51376], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 212, "seek": 33334, "start": 353.55999999999995, "end": 355.91999999999996, "text": "直接通过所有测试用力的比例", "tokens": [51376, 43297, 19550, 16866, 39300, 11038, 233, 5233, 243, 9254, 13486, 1546, 11706, 17797, 51494], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 213, "seek": 33334, "start": 355.91999999999996, "end": 357.79999999999995, "text": "这个指标比PASAT10", "tokens": [51494, 15368, 25922, 162, 3921, 11706, 47, 3160, 2218, 3279, 51588], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 214, "seek": 33334, "start": 357.79999999999995, "end": 359.91999999999996, "text": "尝试十次通过更靠谱", "tokens": [51588, 1530, 251, 5233, 243, 20145, 9487, 19550, 16866, 19002, 5363, 254, 8897, 109, 51694], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 215, "seek": 33334, "start": 359.91999999999996, "end": 361.23999999999995, "text": "因为实际开发中", "tokens": [51694, 34627, 24726, 8842, 227, 18937, 28926, 5975, 51760], "temperature": 0, "avg_logprob": -0.08022285381536834, "compression_ratio": 1.2418879056047198, "no_speech_prob": 1.5427584557081708e-11}, {"id": 216, "seek": 36124, "start": 361.24, "end": 363.38, "text": "你不可能让模型试十次", "tokens": [50365, 2166, 1960, 16657, 33650, 41908, 39823, 5233, 243, 20145, 9487, 50472], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 217, "seek": 36124, "start": 363.38, "end": 365.3, "text": "要的是第一次就尽量对", "tokens": [50472, 4275, 24620, 18049, 9487, 3111, 1530, 121, 26748, 8713, 50568], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 218, "seek": 36124, "start": 365.3, "end": 369.38, "text": "Mindforge 27B的PASAT1是49.51%", "tokens": [50568, 44, 471, 2994, 432, 7634, 33, 1546, 47, 3160, 2218, 16, 1541, 14938, 13, 18682, 4, 50772], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 219, "seek": 36124, "start": 369.38, "end": 371.32, "text": "意味着它几乎一半的任务", "tokens": [50772, 9042, 17268, 20708, 11284, 6336, 254, 2930, 236, 2257, 30018, 1546, 26443, 5087, 94, 50869], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 220, "seek": 36124, "start": 371.32, "end": 373.52, "text": "第一次生成的代码就能跑通", "tokens": [50869, 18049, 9487, 8244, 11336, 1546, 19105, 23230, 223, 3111, 8225, 32585, 19550, 50979], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 221, "seek": 36124, "start": 373.52, "end": 376.92, "text": "而Deepseek V4 Pro用了海量代码数据训练", "tokens": [50979, 11070, 11089, 595, 405, 916, 691, 19, 1705, 9254, 2289, 20078, 26748, 19105, 23230, 223, 33188, 26075, 106, 7422, 255, 10115, 225, 51149], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 222, "seek": 36124, "start": 376.92, "end": 377.94, "text": "参数更大", "tokens": [51149, 2129, 224, 33188, 19002, 3582, 51200], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 223, "seek": 36124, "start": 377.94, "end": 380.8, "text": "但PASAT1只有47.80%", "tokens": [51200, 8395, 47, 3160, 2218, 16, 35244, 14060, 13, 4702, 4, 51343], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 224, "seek": 36124, "start": 380.8, "end": 383.76, "text": "Claude Opus 4.7虽然是顶尖模型", "tokens": [51343, 34, 875, 2303, 12011, 301, 1017, 13, 22, 12026, 121, 5823, 1541, 10178, 114, 1530, 244, 41908, 39823, 51491], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 225, "seek": 36124, "start": 383.76, "end": 386.28000000000003, "text": "但也只比它高1.87个百分点", "tokens": [51491, 8395, 6404, 14003, 11706, 11284, 12979, 16, 13, 23853, 7549, 31906, 6627, 12579, 51617], "temperature": 0, "avg_logprob": -0.10865101593219681, "compression_ratio": 1.083623693379791, "no_speech_prob": 1.6565148558411735e-11}, {"id": 226, "seek": 38628, "start": 386.28, "end": 387.03999999999996, "text": "这说明啥", "tokens": [50365, 5562, 8090, 11100, 3284, 98, 50403], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 227, "seek": 38628, "start": 387.03999999999996, "end": 390.17999999999995, "text": "说明在编程这种强工程属性的任务里", "tokens": [50403, 8090, 11100, 3581, 38109, 244, 29649, 5562, 39810, 5702, 118, 23323, 29649, 9636, 252, 21686, 1546, 26443, 5087, 94, 15759, 50560], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 228, "seek": 38628, "start": 390.17999999999995, "end": 391.38, "text": "知道怎么开发", "tokens": [50560, 7758, 15282, 18937, 28926, 50620], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 229, "seek": 38628, "start": 391.38, "end": 393.71999999999997, "text": "比知道多少代码片段更重要", "tokens": [50620, 11706, 7758, 41492, 19105, 23230, 223, 16668, 28427, 19002, 24928, 50737], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 230, "seek": 38628, "start": 393.71999999999997, "end": 394.76, "text": "更有意思的是", "tokens": [50737, 19002, 2412, 16697, 24620, 50789], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 231, "seek": 38628, "start": 394.76, "end": 396.67999999999995, "text": "研究团队做了个对比实验", "tokens": [50789, 23230, 242, 44704, 3919, 95, 10034, 253, 10907, 2289, 7549, 8713, 11706, 24726, 49657, 234, 50885], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 232, "seek": 38628, "start": 396.67999999999995, "end": 398.53999999999996, "text": "用同样的27B模型", "tokens": [50885, 9254, 13089, 14496, 1546, 10076, 33, 41908, 39823, 50978], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 233, "seek": 38628, "start": 398.53999999999996, "end": 401.28, "text": "一组用1001条完整轨迹训练", "tokens": [50978, 2257, 10115, 226, 9254, 6879, 16, 48837, 14128, 27662, 17819, 101, 3316, 117, 7422, 255, 10115, 225, 51115], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 234, "seek": 38628, "start": 401.28, "end": 403.73999999999995, "text": "另一组用100万条代码片段", "tokens": [51115, 22762, 2257, 10115, 226, 9254, 6879, 23570, 48837, 19105, 23230, 223, 16668, 28427, 51238], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 235, "seek": 38628, "start": 403.73999999999995, "end": 405.67999999999995, "text": "相当于传统数据量训练", "tokens": [51238, 15106, 16233, 37732, 7384, 254, 10115, 253, 33188, 26075, 106, 26748, 7422, 255, 10115, 225, 51335], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 236, "seek": 38628, "start": 405.67999999999995, "end": 406.32, "text": "结果", "tokens": [51335, 45641, 9319, 51367], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 237, "seek": 38628, "start": 406.32, "end": 407.67999999999995, "text": "轨迹训练的模型", "tokens": [51367, 17819, 101, 3316, 117, 7422, 255, 10115, 225, 1546, 41908, 39823, 51435], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 238, "seek": 38628, "start": 407.67999999999995, "end": 409.14, "text": "在Program Bench上", "tokens": [51435, 3581, 12681, 1342, 3964, 339, 5708, 51508], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 239, "seek": 38628, "start": 409.14, "end": 411.41999999999996, "text": "PASAT1是49.51%", "tokens": [51508, 47, 3160, 2218, 16, 1541, 14938, 13, 18682, 4, 51622], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 240, "seek": 38628, "start": 411.41999999999996, "end": 413.59999999999997, "text": "而海量片段训练的模型", "tokens": [51622, 11070, 20078, 26748, 16668, 28427, 7422, 255, 10115, 225, 1546, 41908, 39823, 51731], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 241, "seek": 38628, "start": 413.59999999999997, "end": 415.41999999999996, "text": "只有32.7%", "tokens": [51731, 35244, 11440, 13, 22, 4, 51822], "temperature": 0, "avg_logprob": -0.08772536829898232, "compression_ratio": 1.25, "no_speech_prob": 1.4515293481065505e-11}, {"id": 242, "seek": 41542, "start": 415.42, "end": 417.14000000000004, "text": "差了快17个百分点", "tokens": [50365, 21679, 2289, 10251, 7773, 7549, 31906, 6627, 12579, 50451], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 243, "seek": 41542, "start": 417.14000000000004, "end": 418.22, "text": "这直接证明", "tokens": [50451, 5562, 43297, 5233, 223, 11100, 50505], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 244, "seek": 41542, "start": 418.22, "end": 419.98, "text": "高质量的全流程数据", "tokens": [50505, 12979, 18464, 101, 26748, 1546, 11319, 27854, 29649, 33188, 26075, 106, 50593], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 245, "seek": 41542, "start": 419.98, "end": 422.28000000000003, "text": "比海量的局部数据有效得多", "tokens": [50593, 11706, 20078, 26748, 1546, 34703, 13470, 33188, 26075, 106, 2412, 43076, 5916, 6392, 50708], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 246, "seek": 41542, "start": 422.28000000000003, "end": 424.28000000000003, "text": "那这对中小团队意味着啥", "tokens": [50708, 4184, 5562, 8713, 5975, 7322, 3919, 95, 10034, 253, 9042, 17268, 20708, 3284, 98, 50808], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 247, "seek": 41542, "start": 424.28000000000003, "end": 425.38, "text": "以前大家觉得", "tokens": [50808, 44996, 6868, 29992, 50863], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 248, "seek": 41542, "start": 425.38, "end": 427.22, "text": "要做个能写代码的AI", "tokens": [50863, 4275, 10907, 7549, 8225, 5676, 247, 19105, 23230, 223, 1546, 48698, 50955], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 249, "seek": 41542, "start": 427.22, "end": 428.5, "text": "得有千亿参数", "tokens": [50955, 5916, 2412, 20787, 1369, 123, 2129, 224, 33188, 51019], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 250, "seek": 41542, "start": 428.5, "end": 430.18, "text": "得有几十万块GPU", "tokens": [51019, 5916, 2412, 6336, 254, 20145, 23570, 47734, 38, 8115, 51103], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 251, "seek": 41542, "start": 430.18, "end": 432.04, "text": "得收集TB级的数据", "tokens": [51103, 5916, 18681, 26020, 51, 33, 16853, 100, 1546, 33188, 26075, 106, 51196], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 252, "seek": 41542, "start": 432.04, "end": 433.72, "text": "中小团队根本玩不起", "tokens": [51196, 5975, 7322, 3919, 95, 10034, 253, 31337, 8802, 19912, 26451, 51280], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 253, "seek": 41542, "start": 433.72, "end": 436.02000000000004, "text": "但Mindforge 27B告诉我们", "tokens": [51280, 8395, 44, 471, 2994, 432, 7634, 33, 44816, 15003, 51395], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 254, "seek": 41542, "start": 436.02000000000004, "end": 436.46000000000004, "text": "不用", "tokens": [51395, 24384, 51417], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 255, "seek": 41542, "start": 436.46000000000004, "end": 438.16, "text": "你只需要花精力整理", "tokens": [51417, 2166, 14003, 35748, 20127, 34910, 13486, 27662, 13876, 51502], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 256, "seek": 41542, "start": 438.16, "end": 440.62, "text": "几百上千条高质量的开发轨迹", "tokens": [51502, 6336, 254, 31906, 5708, 20787, 48837, 12979, 18464, 101, 26748, 1546, 18937, 28926, 17819, 101, 3316, 117, 51625], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 257, "seek": 41542, "start": 440.62, "end": 442.16, "text": "用个小参数模型", "tokens": [51625, 9254, 7549, 7322, 2129, 224, 33188, 41908, 39823, 51702], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 258, "seek": 41542, "start": 442.16, "end": 443.28000000000003, "text": "比如27B", "tokens": [51702, 36757, 10076, 33, 51758], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 259, "seek": 41542, "start": 443.28000000000003, "end": 445.22, "text": "几张V100卡就能训练", "tokens": [51758, 6336, 254, 44059, 53, 6879, 32681, 3111, 8225, 7422, 255, 10115, 225, 51855], "temperature": 0, "avg_logprob": -0.08078175453684437, "compression_ratio": 1.2267441860465116, "no_speech_prob": 1.3568663553842342e-11}, {"id": 260, "seek": 44522, "start": 445.22, "end": 447.94000000000005, "text": "就能做出接近顶尖大模型的效果", "tokens": [50365, 3111, 8225, 10907, 7781, 14468, 17463, 10178, 114, 1530, 244, 3582, 41908, 39823, 1546, 43076, 9319, 50501], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 261, "seek": 44522, "start": 447.94000000000005, "end": 449.98, "text": "这对资源有限的团队来说", "tokens": [50501, 5562, 8713, 5266, 226, 47402, 2412, 43446, 1546, 3919, 95, 10034, 253, 6912, 8090, 50603], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 262, "seek": 44522, "start": 449.98, "end": 451.12, "text": "简直是福音", "tokens": [50603, 11249, 222, 16186, 1541, 33952, 18034, 50660], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 263, "seek": 44522, "start": 451.12, "end": 452.22, "text": "不用卷算力", "tokens": [50660, 24384, 5322, 115, 19497, 13486, 50715], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 264, "seek": 44522, "start": 452.22, "end": 453.44000000000005, "text": "不用卷数据量", "tokens": [50715, 24384, 5322, 115, 33188, 26075, 106, 26748, 50776], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 265, "seek": 44522, "start": 453.44000000000005, "end": 454.94000000000005, "text": "卷数据质量就行了", "tokens": [50776, 5322, 115, 33188, 26075, 106, 18464, 101, 26748, 3111, 46606, 50851], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 266, "seek": 44522, "start": 454.94000000000005, "end": 456.84000000000003, "text": "比如一个小团队想做一个", "tokens": [50851, 36757, 20182, 7322, 3919, 95, 10034, 253, 7093, 10907, 20182, 50946], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 267, "seek": 44522, "start": 456.84000000000003, "end": 458.78000000000003, "text": "金融代码生成的专业模型", "tokens": [50946, 19117, 43772, 235, 19105, 23230, 223, 8244, 11336, 1546, 940, 241, 940, 248, 41908, 39823, 51043], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 268, "seek": 44522, "start": 458.78000000000003, "end": 461.26000000000005, "text": "他们不用去爬Github上所有代码", "tokens": [51043, 47911, 24384, 6734, 8164, 105, 38, 355, 836, 5708, 39300, 19105, 23230, 223, 51167], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 269, "seek": 44522, "start": 461.26000000000005, "end": 463.62, "text": "只需要找几个资深金融工程师", "tokens": [51167, 14003, 35748, 25085, 6336, 254, 7549, 5266, 226, 24043, 19117, 43772, 235, 23323, 29649, 29186, 51285], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 270, "seek": 44522, "start": 463.62, "end": 465.5, "text": "记录他们开发风控系统", "tokens": [51285, 34756, 7391, 243, 47911, 18937, 28926, 47209, 48707, 25368, 10115, 253, 51379], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 271, "seek": 44522, "start": 465.5, "end": 466.92, "text": "交易接口的全过程", "tokens": [51379, 28455, 31962, 14468, 18144, 1546, 11319, 16866, 29649, 51450], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 272, "seek": 44522, "start": 466.92, "end": 468.5, "text": "整理成几百条轨迹", "tokens": [51450, 27662, 13876, 11336, 6336, 254, 31906, 48837, 17819, 101, 3316, 117, 51529], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 273, "seek": 44522, "start": 468.5, "end": 470.72, "text": "就能训练出一个懂金融业务", "tokens": [51529, 3111, 8225, 7422, 255, 10115, 225, 7781, 20182, 29624, 19117, 43772, 235, 940, 248, 5087, 94, 51640], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 274, "seek": 44522, "start": 470.72, "end": 472.22, "text": "懂工程逻辑的模型", "tokens": [51640, 29624, 23323, 29649, 2215, 119, 9830, 239, 1546, 41908, 39823, 51715], "temperature": 0, "avg_logprob": -0.06613065729189159, "compression_ratio": 1.3533123028391167, "no_speech_prob": 1.404088043749363e-11}, {"id": 275, "seek": 47222, "start": 472.22, "end": 473.94000000000005, "text": "这种模型虽然参数小", "tokens": [50365, 5562, 39810, 41908, 39823, 12026, 121, 5823, 2129, 224, 33188, 7322, 50451], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 276, "seek": 47222, "start": 473.94000000000005, "end": 476.02000000000004, "text": "但比那些啥都懂一点的大模型", "tokens": [50451, 8395, 11706, 4184, 13824, 3284, 98, 7182, 29624, 49042, 1546, 3582, 41908, 39823, 50555], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 277, "seek": 47222, "start": 476.02000000000004, "end": 478.16, "text": "在实际金融场景里更靠谱", "tokens": [50555, 3581, 24726, 8842, 227, 19117, 43772, 235, 50255, 50218, 15759, 19002, 5363, 254, 8897, 109, 50662], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 278, "seek": 47222, "start": 478.16, "end": 479.24, "text": "最后总结一下", "tokens": [50662, 8661, 13547, 33440, 45641, 8861, 50716], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 279, "seek": 47222, "start": 479.24, "end": 481.3, "text": "Mindforge 27B的成功", "tokens": [50716, 44, 471, 2994, 432, 7634, 33, 1546, 11336, 32311, 50819], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 280, "seek": 47222, "start": 481.3, "end": 483.82000000000005, "text": "其实打破了一个长期存在的迷思", "tokens": [50819, 41646, 12467, 39560, 2289, 20182, 32271, 16786, 39781, 3581, 1546, 3316, 115, 8870, 50945], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 281, "seek": 47222, "start": 483.82000000000005, "end": 485.86, "text": "编程能力等于参数规模", "tokens": [50945, 38109, 244, 29649, 8225, 13486, 10187, 37732, 2129, 224, 33188, 6758, 226, 41908, 51047], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 282, "seek": 47222, "start": 485.86, "end": 486.62, "text": "它证明", "tokens": [51047, 11284, 5233, 223, 11100, 51085], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 283, "seek": 47222, "start": 486.62, "end": 488.66, "text": "在编程这种需要强逻辑", "tokens": [51085, 3581, 38109, 244, 29649, 5562, 39810, 35748, 5702, 118, 2215, 119, 9830, 239, 51187], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 284, "seek": 47222, "start": 488.66, "end": 490.3, "text": "强工程思维的领域", "tokens": [51187, 5702, 118, 23323, 29649, 8870, 10115, 112, 1546, 12501, 228, 16262, 253, 51269], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 285, "seek": 47222, "start": 490.3, "end": 491.36, "text": "数据的质量", "tokens": [51269, 33188, 26075, 106, 1546, 18464, 101, 26748, 51322], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 286, "seek": 47222, "start": 491.36, "end": 492.58000000000004, "text": "是不是全流程", "tokens": [51322, 23034, 11319, 27854, 29649, 51383], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 287, "seek": 47222, "start": 492.58000000000004, "end": 494.42, "text": "是不是包含试错过程", "tokens": [51383, 23034, 23305, 2392, 104, 5233, 243, 29900, 16866, 29649, 51475], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 288, "seek": 47222, "start": 494.42, "end": 496.16, "text": "是不是有完整上下文", "tokens": [51475, 23034, 2412, 14128, 27662, 5708, 4438, 17174, 51562], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 289, "seek": 47222, "start": 496.16, "end": 497.56, "text": "比数据量更重要", "tokens": [51562, 11706, 33188, 26075, 106, 26748, 19002, 24928, 51632], "temperature": 0, "avg_logprob": -0.08289365110726192, "compression_ratio": 1.244299674267101, "no_speech_prob": 1.5315327131504297e-11}, {"id": 290, "seek": 49756, "start": 497.56, "end": 498.56, "text": "就像教徒弟", "tokens": [50365, 3111, 12760, 21936, 2172, 240, 19630, 50415], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 291, "seek": 49756, "start": 498.56, "end": 500.64, "text": "你让他看一万行代码片段", "tokens": [50415, 2166, 33650, 5000, 4200, 2257, 23570, 8082, 19105, 23230, 223, 16668, 28427, 50519], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 292, "seek": 49756, "start": 500.64, "end": 503.06, "text": "不如让他跟着你完整做十个项目", "tokens": [50519, 1960, 8238, 33650, 5000, 9678, 20708, 2166, 14128, 27662, 10907, 20145, 7549, 10178, 117, 11386, 50640], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 293, "seek": 49756, "start": 503.06, "end": 505.28000000000003, "text": "因为前者只能让他记住怎么写", "tokens": [50640, 34627, 8945, 12444, 14003, 8225, 33650, 5000, 34756, 21632, 15282, 5676, 247, 50751], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 294, "seek": 49756, "start": 505.28000000000003, "end": 507.22, "text": "后者能让他学会怎么做", "tokens": [50751, 13547, 12444, 8225, 33650, 5000, 29618, 12949, 15282, 10907, 50848], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 295, "seek": 49756, "start": 507.22, "end": 507.78000000000003, "text": "未来", "tokens": [50848, 29954, 6912, 50876], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 296, "seek": 49756, "start": 507.78000000000003, "end": 510.96, "text": "可能会有更多这种小而美的专业模型出现", "tokens": [50876, 16657, 12949, 2412, 19002, 6392, 5562, 39810, 7322, 11070, 9175, 1546, 940, 241, 940, 248, 41908, 39823, 7781, 20204, 51035], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 297, "seek": 49756, "start": 510.96, "end": 512.64, "text": "他们不需要千亿参数", "tokens": [51035, 47911, 1960, 35748, 20787, 1369, 123, 2129, 224, 33188, 51119], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 298, "seek": 49756, "start": 512.64, "end": 515.24, "text": "只需要几千条高质量的领域轨迹", "tokens": [51119, 14003, 35748, 6336, 254, 20787, 48837, 12979, 18464, 101, 26748, 1546, 12501, 228, 16262, 253, 17819, 101, 3316, 117, 51249], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 299, "seek": 49756, "start": 515.24, "end": 517.84, "text": "就能在特定领域达到顶尖水平", "tokens": [51249, 3111, 8225, 3581, 17682, 12088, 12501, 228, 16262, 253, 9830, 122, 4511, 10178, 114, 1530, 244, 15590, 16716, 51379], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 300, "seek": 49756, "start": 517.84, "end": 520.16, "text": "这对AI的普及来说是件好事", "tokens": [51379, 5562, 8713, 48698, 1546, 29993, 25703, 6912, 8090, 1541, 20485, 2131, 6973, 51495], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 301, "seek": 49756, "start": 520.16, "end": 522.2, "text": "毕竟不是每个团队都有谷歌", "tokens": [51495, 7256, 243, 11957, 253, 7296, 23664, 7549, 3919, 95, 10034, 253, 48121, 8897, 115, 29582, 51597], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 302, "seek": 49756, "start": 522.2, "end": 523.48, "text": "OpenAI的资源", "tokens": [51597, 45569, 48698, 1546, 5266, 226, 47402, 51661], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 303, "seek": 49756, "start": 523.48, "end": 526.58, "text": "但每个团队都可能拥有高质量的领域数据", "tokens": [51661, 8395, 23664, 7549, 3919, 95, 10034, 253, 7182, 16657, 6852, 98, 2412, 12979, 18464, 101, 26748, 1546, 12501, 228, 16262, 253, 33188, 26075, 106, 51816], "temperature": 0, "avg_logprob": -0.061012671544001654, "compression_ratio": 1.3735294117647059, "no_speech_prob": 1.2411887490015872e-11}, {"id": 304, "seek": 52658, "start": 526.58, "end": 529.7, "text": "Mindforge 27B算是给我们开了个好投", "tokens": [50365, 44, 471, 2994, 432, 7634, 33, 19497, 1541, 23197, 15003, 18937, 2289, 7549, 2131, 37094, 50521], "temperature": 0, "avg_logprob": -0.15343785285949707, "compression_ratio": 0.7962962962962963, "no_speech_prob": 1.9753927660293158e-11}], "language": "zh"}