NVIDIA Launched its own version ChatGPT, The name is very GPU taste——
ChatWithRTX。
This one from NVIDIA AI Chatbots are different from current mainstream players. It is not on a web page or APP It needs to be downloaded and installed on a personal computer. This wave operate , Not only will it be faster in terms of running results, it also means ChatWithRTX There may not be so many limits on chat content.
Netizens also lamented this point.：
Of course, in terms of performance aspect It is also required, only at least 8 GB of RTX30 Or 40 series graphics card.
So ChatWithRTX Let’s continue to look at the practical results.
First of all, it is worth mentioning that ChatWithRTX It’s not that Nvidia has built a big language model ( LLM）. It still relies on two open source LLM, Right now Mistral and Llama2, User is running time You can choose according to your preference.
Pick over LLM After that, you can ChatWithRTX Upstream local files. Allowed file types include txt，.pdf，.doc/.docx and. xml。
Then you can start asking questions, such as：
Because it is running locally, so ChatWithRTX generating answers speed It was extremely fast, a real snap.：
besides, ChatWithRTX Another highlight feature is that it can be based on online video Make an answer. For example, feed it a link to a YouTube video:
and backward ChatWithRTX Ask a question：
ChatWithRTX They will also answer questions based on the video content very quickly. As for what is used in front of it technology On the other hand, Nvidia’s official website only mentions that it is difficult to: Retrieval enhancement generation is used ( RAG）、NVIDIATensorRTLLM software and NVIDIARTX wait.
As we just mentioned, ChatWithRTX The usage is very complicated, there is only one download and installation method. However, in terms of performance, in addition to GPU In addition to the requirements, there are also some conditions, such as：
system： Windows10 or Windows11
RAM: at least 16 GB
drive: Version 535.11 or downgrade
but ChatWithRTX Not very lightweight in terms of size, about 35 in total G. So before downloading it, be sure to review it ChatWithRTX Required installation conditions. Otherwise, there will be all kinds of comedies.:
TheVerge Published at NVIDIA ChatWithRTX After that, a wave of actual tests was immediately carried out. But the conclusion was shocking. For example, the video search performance just mentioned, during the actual test session, it actually downloaded text records of completely different videos.
Secondly, suppose that given ChatWithRTX Feeding too many files, such as letting ChatwithRTX for 25,000 document Indexing, it indirectly strikes disintegration. And it can't remember the context, so subsequent questions can't be based on the previous question.
Finally, there is a drawback, which is downloading ChatwithRTX It took the tester a full half an hour...but aside from the shortcomings, TheVerge It is also more necessary than a neutral ground. ChatwithRTX benefits. For example, the speed and accuracy of searching documents on a computer are amazing.
And also make a summary ChatwithRTX Longer content：
Even key What's more, running such a mechanism locally guarantees the security of user files. then you will pick NVIDIA version ChatGPT ?
This article comes from the WeChat public account Qubits ( ID：QbitAI）, author: Paying attention to cutting-edge technology, 36 Krypton was authorized and issued.
