The top domestic large-scale model "Xunfei Spark": image generation, code generation, support for plug-ins, etc.

On August 12, in the "Artificial Intelligence Large Model Experience Report 2.0" report released by the China Enterprise Development Research Center of the Xinhua News Agency Research Institute, Xunfei Xinghuo was rated as one of the domestic excellent large models with a total score of 1013 points. (Experience address: https://xinghuo.xfyun.cn/?ch=8tcbd7p)

insert image description here

Xunfei Xinghuo can provide multi-scene text generation, language understanding, knowledge question and answer, logical reasoning, mathematical answering, code generation and multi-modal capabilities based on natural text and voice, and quickly generate text, pictures, codes and other content. Applicable to all mainstream business scenarios such as marketing, e-commerce, finance, recruitment, customer service, medical care, and copywriting.

From the point of view of Chinese creation alone, the text generated by Xunfei Xinghuo can better meet the domestic needs. In terms of generation speed, coherence, text simplification, and text expression, it also shows unique Chinese advantages compared with ChatGPT and other products. After the whole experience, it belongs to the best among domestic large-scale models.

In terms of multi-modality, Xunfei Xinghuo has also made a new update this time, supporting the function of asking questions by pictures, which is currently not available on GPT-4. Users can send a picture to Xunfei Xinghuo, and then ask Xunfei Xinghuo questions based on the content of the picture, for example, send a picture of the Big Wild Goose Pagoda in Xi'an, and ask it to identify which building the picture shows and what historical allusions are there wait.

In addition, in this update, Xunfei Spark also provides a visual "data fine-tuning" feature, which supports users to upload 10 .txt and .PDF texts at a time, so that Xunfei Xinghuo can answer content in specific fields. For example, if you upload 5 legal documents, you can ask about the contents of the documents.

The following "AIGC Open Community" will give you an in-depth interpretation of the main features of Xunfei Spark:

  1. Image generation: According to the user's description, generate an image that meets the requirements.

  2. Image understanding: Support user image input, and perform visual questions and answers on image content.

  3. Virtual human video generation: According to user requirements, generate a virtual human video that meets the requirements.

  4. Code ability: Provides powerful code generation, interpretation, and error correction functions.

  5. Assistant function upgrade: 1) Help users quickly customize their own exclusive assistant through fields such as role setting, goal task, requirement description, style setting, and improve the effect of the assistant's large-scale application. 2) Personal datasets: Users can create datasets and link them to the assistant for question-and-answer. 3) Independent web application: A single assistant can be shared as an independent web application. 4) Assistant API: Each assistant automatically generates an independent API to help developers quickly integrate the assistant into their own applications.

  6. Plug-in function: Three powerful plug-in functions can be selected in the dialog box, document question and answer, PPT generation and resume generation.

text generated image

Xunfei Xinghuo is a multi-modal output model that supports text to generate images directly. Although this function is not as delicate as Midjourney, it can basically understand the user's intentions and generate high-quality pictures. For example, spawn an off-road vehicle racing through the desert.

insert image description here

Generate a sharp blade that cuts iron like mud

insert image description here

Generate a fat doll holding a big watermelon, two-dimensional style

insert image description here

picture understanding

Picture questioning is one of the main features of Xunfei Xinghuo. As early as March this year, when OpenAI released GPT-4, it showed that this function has not been opened until now. Now, you can experience it on Xunfei Xinghuo.

1) Uploaded a picture of the Great Wall, quickly recognized the building in the picture and introduced its history.

insert image description here

2) Upload an anime picture, and quickly identify the picture elements and give a detailed interpretation. When asked which anime it was, Xunfei Xinghuo actually answered "Naruto" accurately, which is quite surprising.

insert image description here

3) In order to test the deep recognition function of Xunfei Xinghuo, I input a computer picture with garbled characters, and it can still be recognized and give an accurate answer.

insert image description here

Virtual Human Video Generation

This function can generate a virtual human video that meets the requirements according to the text prompt. For example, help me generate a video to wish Mrs. Li, the enlightenment mentor, a happy Teacher's Day. After getting the copy, click Generate Video to complete.

insert image description here

HKUST Xunfei Spark

Video display generated by Xunfei Xinghuo

If the user is dissatisfied with the image, background, and sound, he can click Customize to replace it and select elements that meet the text requirements.

insert image description here

Multi-scene text generation

In order to meet the needs of different users, Xunfei Xinghuo provides an assistant center that can be used in rich scenarios, which can provide more in-depth and professional text creative functions according to fields such as workplace, marketing, life, recruitment, programming, and customer service.

Including title generation, mock interview, story creation, scenic spot recommendation, etc., suitable for Xiaohongshu, Douyin, Weibo, Kuaishou and other platforms, it is very convenient to use just by clicking.

insert image description here

For the title of Xiaohongshu, directly enter the product name, for example, hydrating mask, fitness clothes, and three popular titles can be generated immediately. If you are not satisfied, click to answer again.

insert image description here

insert image description here

PPT outline assistant, the user only needs to input what you want to do, for example, enter a weight loss plan, and the detailed content of the PPT can be generated, including cover, introduction, goal, diet plan, habit adjustment, etc.

insert image description here

For job application emails, users only need to enter the position they want to apply for, and a comprehensive cover letter can be generated. For example, Marketing Director.

insert image description here

For text workers, proofreading of manuscripts is laborious and laborious, and it is not easy to find subtle errors and differences. Xunfei Xinghuo provides professional manuscript proofreading assistants.

For example, I went to a big library yesterday, there are many books. I saw an interesting book and I ate it. Typos and grammatical errors in the statement can be identified and a revised answer given.

insert image description here

The extension assistant is very helpful for the creation of copywriting. It only needs a paragraph or even a few words to describe a richer scene. For example, it rained heavily in the sky.

insert image description here

The assistant center provided by Xunfei Xinghuo is very rich, covering almost all business scenarios. This function is mainly for C-end users, which lowers the threshold for users to use large models. It does not need to master complex prompt word engineering knowledge to apply to large models. Conduct professional questioning interactions to obtain accurate answers.

Code generation, interpretation, error correction functions

Xunfei Spark provides powerful code generation, interpretation, and error correction functions. For example, using Java to write a login page for a travel website, a simple example is quickly given, and the operating principle, calling, and operating principle of the code are explained in detail.

insert image description here

Enter a piece of Python code and ask it to explain how it works, what it does, etc.

insert image description here

Entering a piece of wrong Python code can quickly identify and propose a detailed modification plan, while generating a correct code example.

insert image description here

Powerful plug-in function

Users can choose three powerful plug-in functions in the dialog box, document question and answer, PPT generation and resume generation, among which PPT and resume generation support document download, and the function is quite powerful.

insert image description here

Document question and answer, upload an AIGC in-depth report, and then start asking questions based on the content of the document. For example, sorting out the beneficiary manufacturers of AIGC, the content of the answers is basically consistent with the uploaded documents, and the recognition rate and accuracy of the answers are very high.

insert image description here

insert image description here

Screenshot of source document content

PPT template generation, for example, to generate a sales PPT template, users can directly download the generated PPT template on the platform.

insert image description here

Resume generation, for example, my name is Zhang Sanfeng, I have five years of real estate sales experience, help me generate a resume. The template can be downloaded in Word format.

insert image description here

Create a custom helper function

Users can create specific chat assistants based on their own data, such as financial, sales, education, and historical data.

This function is equivalent to the visual "data fine-tuning" function, and users can also publish their own assistants to the platform for others to use.

The creation process is very simple. The user logs in to https://xinghuo.xfyun.cn/botcenter and clicks Create Assistant, enters the assistant's name, marketing, function introduction, etc., and clicks Create.

insert image description here

Then click the data set to add professional data to the created assistant, and support users to upload 10 .txt and PDF files at a time, with no more than 200,000 words in a single file.

insert image description here

After the data upload is complete, click the created assistant and associate the dataset we just created in the knowledge base source to complete. Now, you can ask questions according to the custom-created assistant and upload the document content of the data.

More detailed creation tutorials can be viewed at https://developer.xfyun.cn/thread/116465.

HKUST Xunfei will hold the "Xunfei Spark Cognitive Model V2.0" upgrade conference at 14:00 today. The focus of this conference will be on the significant improvements in code capabilities and multimodal capabilities brought about by this major version upgrade, and the launch of more diverse scenario-based products for teachers, schools, enterprises, developers, etc.

The theme of the press conference is "Liberating Productivity, Unleashing Imagination". Interested partners can go to the official video account of Xunfei Open Platform to make an appointment or watch it on Xinghuo's official website~

Guess you like

Origin blog.csdn.net/weixin_57291105/article/details/132297499