webAI发布1.7B逻辑专家模型TwiL-LM3,手机可跑且推理超更大模型
webAI built a 1-gigabyte logic expert open-source model that runs on a phone and…
webAI built a 1-gigabyte logic expert open-source model that runs on a phone and out-reasons models more than twice its size
webAI 构建了一个1GB的逻辑专家开源模型,可在手机上运行,其推理能力超过两倍于其规模的模型。
TwiL-LM3, a 1.7B formal-logic model that beats OpenAI's gpt-oss-120b on 4 of 5 formal reasoning benchmarks while running efficiently on consumer hardware. Available on Huggingface.
TwiL-LM3,一个1.7B的正式逻辑模型,在5个正式推理基准测试中的4个上击败了OpenAI的gpt-oss-120b,同时在消费级硬件上高效运行。可在Huggingface上获取。
That's 40× fewer parameters, 2.6× faster inference
参数减少了40倍,推理速度提高了2.6倍。
The model does one specialized task, turning plain English into formal logic and checking whether conclusions follow from premises.
该模型执行一项专门任务,将普通英语转换为正式逻辑,并检查结论是否从前提中得出。
i.e. taking statements written in ordinary language and working out, step by step, what does and does not follow from them.
即,将用日常语言书写的陈述,逐步推导出哪些结论成立,哪些不成立。
Its trained on a purpose-engineered, proprietary formal-logic data engine built from open sources.
它是在一个专门设计、专有的正式逻辑数据引擎上训练的,该引擎基于开源资源构建。
Instead of training from scratch, webAI fine-tuned base models with a 289MB LoRA adapter of roughly 72M parameters.
webAI 并非从头训练,而是使用一个约7200万参数的289MB LoRA适配器对基础模型进行了微调。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力