Article Directory
product definition
Wenxin Qianfan large-scale model platform is a one-stop large-scale model development and service operation platform for enterprise developers. Wenxin Qianfan not only provides the underlying model including Wenxin Yiyan (ERNIE-Bot) and third-party open source large models, but also provides various AI development tools and a complete set of development environment to facilitate customers to easily use and develop large model applications.
Wenxin Qianfan provides one-stop large-scale model customization services for data management, automated model SFT, and inference service cloud deployment to help meet the needs of generative AI applications in various industries.
Application Scenario
To meet the needs of different enterprises, Wenxin Qianfan provides different functional services.
For example, for general application scenarios such as intelligent dialogue and intelligent input method, Wenxin Yiyan enterprise-level reasoning cloud service can be directly experienced in Wenxin Qianfan, and business integration can be carried out;
Product introductions and promotional articles in subdivided telemarketing scenarios need to customize and adjust the reasoning results. You can use the online fine-tuning training function of the model in Wenxin Qianfan to quickly produce customized model services for industry scenarios to meet the specific needs of users.
Scenario 1: Dialogue and Communication
Matching scenarios: intelligent marketing, intelligent customer service, emotional communication and other scenarios that require communication and dialogue.
In real life, provide quick responses to user needs, accurately match user needs, complete marketing business development, timely response, positive psychological counseling, etc., to improve customer experience.
Specific case: output results according to user needs. For example, users need the platform to help recommend the four major classics, as shown in the following figure:
Scenario 2: Content Creation
Matching scenarios: text creation scenarios such as scripts, stories, and poems.
According to the user's needs, generate accurately matched creative texts, provide users with script sources for video editing; polish and shape stories, poems and other text content, and create a cultural environment for users to improve their text capabilities.
Specific case: users issue custom instructions to create shaped text content. For example, the user needs the platform to write a Tibetan acrostic poem as required, as shown in the following figure:
Scenario 3: Analysis and Control
Matching scenario: The so-called analysis control is a text scenario that includes deep learning such as code generation, data reporting, and content analysis.
Quickly generate executable code according to the user's needs or according to the user's needs, the platform combines various data it already has to generate a response content with a higher degree of matching.
Specific case: When a user temporarily encounters a problem that needs to be dealt with, the platform generates a solution. For example, development engineers use the platform to generate specific code and complete code optimization, as shown in the following figure:
other scenes
- Government services: multiple rounds of intelligent interaction, bringing warmth to people's livelihood services.
-
Financial Services: Ability comparable to professionals, assisting scientific financial decision-making.
-
Travel service: One-stop itinerary planning and booking, each trip is customized.
- E-commerce service: Automatically generate delivery scripts, digital human anchors start broadcasting with one click.
-
Enterprise office-PPT production: Say goodbye to repeated format adjustments, and make professional PPTs in minutes.
-
Enterprise Office - Graphite Documents: Easily complete various types of writing and unleash unlimited creativity.
How to use Wenxin Qianfan
Wenxin Qianfan provides flexible deployment methods for enterprise customers. There are out-of-the-box, simple and efficient public cloud services, as well as privatized deployment methods, so that customers with high compliance requirements can also enjoy the dividends of large-scale models.
Wenxin Qianfan product advantages
-
Strong foundation and rich knowledge: Wenxin Qianfan platform is based on Baidu Smart Cloud, adopts the deep learning framework of Flying Paddle as the underlying support, and has built-in Wenxin large-scale model technology. Users can easily obtain high-precision and high-performance large models through a small amount of data adjustment.
-
Perfect process and convenient release: provide one-stop service, covering data set management, model training, service release and supervision. Realize the full lifecycle management of models through a visual interface, simplify the implementation process of large models from data to services, and are easy to use and understand.
-
Stable operation and co-construction of ecology: Wenxin Qianfan has a complete technology stack, long-term stable model development engine and excellent performance. The platform has low technical barriers and is suitable for access from all walks of life, helping to complete the development and construction of large-scale industry models.
-
Safe and reliable, one-click activation: Wenxin Qianfan provides Wenxin Yiyan enterprise-level services, combined with Baidu's smart cloud security control mechanism and Wenxin's underlying content security functions, to review reasoning content and filter sensitive words to ensure safety and credibility.
Application process
Apply for closed beta
Official link: https://login.bce.baidu.com/. If you don't have a Baidu account, register for one! If there is a direct login!
Apply for the internal test, and then wait for the review, about three hours!
After passing the review, a voucher for testing will be issued!
The link is as follows:
https://console.bce.baidu.com/finance/#/finance/coupon/list
Note:
There are vouchers, and the applicable products include the Wenxin Qianfan large model platform.
The order type must be: General or Postpaid
Open payment
Visit the activation product page
https://console.bce.baidu.com/ai/?#/ai/wenxinworkshop/create/home
Click to use now
Pull the scroll bar all the way down. Then, click Agree and Continue.
click next
click next
Congratulations, it's activated!
Then, on the main page, I saw that the large model of Wenxin Qianfan was included in the ordered products.
Create an application
Click Wenxin Qianfan Large Model Platform to enter the platform.
In the red box below, you can see the service we just opened. Click on the red box above to create an application!
Enter the name, just define the name yourself, and the following introduction, fill in as you like! The default interface already contains the large model, we don't need to modify it!
Then click Create Now.
Congratulations, you have completed the creation!
large model reasoning
There are three ways to use it, including: online test,
Open the link: https://ai.baidu.com/wenxinworkshop/app/overview, the interface is as follows: The
interface in the above picture is the entire functional platform of Wenxin Qianfan, including testing, training, and data set management.
Click Test Online and the conversation is ready to begin. On the right is the model and the app we just created!
First send a proposition!
Regarding the question of Lu Xun and Zhou Shuren, many big models have answered wrongly! This didn't bother Wenxin Qianfan!
The picture above is the result of Llama2 70B Chatbot's answer, not only wrong, but also in English! The support for Chinese is not very good!
Next question:
It's hard to fail! Another question:
This model is a bit fierce!
release deployment
Click Service Management to enter the release page,
select the deployment method, model, model version, and fill in the service name and interface address. Then, click Next!
Choose a time and apply for publication! However, so expensive! ! ! Be sure to touch your pocket before ordering!
data set
Make a dataset
[{
"prompt": "类型#上衣*材质#牛仔布*颜色#白色*风格#简约*图案#刺绣*衣样式#外套*衣款式#破洞", "response": [["简约而不简单的牛仔外套，白色的衣身十分百搭。衣身多处有做旧破洞设计，打破单调乏味，增加一丝造型看点。衣身后背处有趣味刺绣装饰，丰富层次感，彰显别样时尚。"]]}]
[{
"prompt": "类型#裙*材质#针织*颜色#纯色*风格#复古*风格#文艺*风格#简约*图案#格子*图案#纯色*图案#复古*裙型#背带裙*裙长#连衣裙*裙领型#半高领", "response": [["这款BRAND针织两件套连衣裙，简约的纯色半高领针织上衣，修饰着颈部线，尽显优雅气质。同时搭配叠穿起一条背带式的复古格纹裙，整体散发着一股怀旧的时髦魅力，很是文艺范。"]]}]
[{
"prompt": "类型#上衣*风格#嘻哈*图案#卡通*图案#印花*图案#撞色*衣样式#卫衣*衣款式#连帽", "response": [["嘻哈玩转童年，随时<UNK>，没错，出街还是要靠卫衣来装酷哦！时尚个性的连帽设计，率性有范还防风保暖。还有胸前撞色的卡通印花设计，靓丽抢眼更富有趣味性，加上前幅大容量又时尚美观的袋鼠兜，简直就是孩子耍帅装酷必备的利器。"]]}]
[{
"prompt": "类型#裤*风格#英伦*风格#简约", "response": [["裤子是简约大方的版型设计，带来一种极简主义风格而且不乏舒适优雅感，是衣橱必不可少的一件百搭单品。标志性的logo可以体现出一股子浓郁的英伦风情，轻而易举带来独一无二的<UNK>体验。"]]}]
[{
"prompt": "类型#裙*裙下摆#弧形*裙腰型#高腰*裙长#半身裙*裙款式#不规则*裙款式#收腰", "response": [["这款来自梵凯的半身裙富有十足的设计感，采用了别致的不规则设计，凸显出时尚前卫的格调，再搭配俏皮的高腰设计，收腰提臀的同时还勾勒出优美迷人的身材曲线，而且还帮你拉长腿部比例，释放出优雅娇俏的小女人味。并且独特的弧形下摆还富有流畅的线条美，一颦一动间展现出灵动柔美的气质。"]]}]
[{
"prompt": "类型#上衣*版型#宽松*版型#显瘦*图案#线条*衣样式#衬衫*衣袖型#泡泡袖*衣款式#抽绳", "response": [["这件衬衫的款式非常的宽松，利落的线条可以很好的隐藏身材上的小缺点，穿在身上有着很好的显瘦效果。领口装饰了一个可爱的抽绳，漂亮的绳结展现出了十足的个性，配合时尚的泡泡袖型，尽显女性甜美可爱的气息。"]]}]
[{
"prompt": "类型#裙*材质#蕾丝*风格#宫廷*图案#刺绣*图案#蕾丝*裙型#大裙摆*裙下摆#花边*裙袖型#泡泡袖", "response": [["宫廷风的甜美蕾丝设计，清醒的蕾丝拼缝处，刺绣定制的贝壳花边，增添了裙子的精致感觉。超大的裙摆，加上精细的小花边设计，上身后既带着仙气撩人又很有女人味。泡泡袖上的提花面料，在细节处增加了浪漫感，春日的仙女姐姐。浪漫蕾丝布满整个裙身，美丽明艳，气质超仙。"]]}]
[{
"prompt": "类型#裤*版型#显瘦*颜色#黑色*风格#简约*裤长#九分裤", "response": [["个性化的九分裤型，穿着在身上，能够从视觉上拉长你的身体比例，让你看起来更加的有范。简约的黑色系列，极具时尚的韵味，充分凸显你专属的成熟韵味。修身的立体廓形，为你塑造修长的曲线。"]]}]
[{
"prompt": "类型#裙*版型#显瘦*风格#文艺*风格#简约*图案#印花*图案#撞色*裙下摆#压褶*裙长#连衣裙*裙领型#圆领", "response": [["文艺个性的印花连衣裙，藏青色底蕴，低调又大气，撞色太阳花分布整个裙身，绚丽而美好，带来时尚减龄的气质。基础款的舒适圆领，简约不失大方，勾勒精致脸庞。领后是一粒包布扣固定，穿脱十分方便。前片立体的打褶设计，搭配后片压褶的做工，增添层次和空间感，显瘦又有型。"]]}]
[{
"prompt": "类型#裙*颜色#蓝色*风格#清新*图案#蝴蝶结", "response": [["裙身处采用立体蝴蝶结装饰辅以蓝色条带点缀，令衣身造型饱满富有层次的同时为其注入一丝甜美气息。将女孩清新娇俏的一面衬托而出。"]]}]
[{
"prompt": "类型#裙*颜色#白色*风格#清新*图案#碎花*裙腰型#松紧腰*裙长#长裙*裙衣门襟#拉链*裙款式#拉链", "response": [["这条颜色素雅的长裙，以纯净的白色作为底色，辅以印在裙上的点点小碎花，<UNK>勾勒出一幅生动优美的“风景图”，给人一种大自然的清新之感，好似吸收新鲜空气的那种舒畅感。腰间贴心地设计成松紧腰，将腰线很好地展现出来，十分纤巧，在裙子的侧边，有着一个隐形的拉链，能够让你穿脱自如。"]]}]
[{
"prompt": "类型#裤*材质#羊毛*裤长#九分裤*裤口#微喇裤", "response": [["不同于一般的西服裤。这款<UNK>小喇叭羊毛裤在样式上显得更加时髦优雅，特地采用微微的九分喇叭裤腿设计，视觉上将脚踝处显得更加纤细。并且特地甄选柔软的羊毛材质，就算直接贴肤穿着，也不会觉得寒冷，比较适合初秋穿噢。"]]}]
[{
"prompt": "类型#上衣*风格#简约*衣门襟#拉链*衣款式#口袋*衣款式#拉链", "response": [["上衣与裤子的连体式设计从整体看起来十分的具有大牌的风范。简约，没有任何的其他装饰，把自己的独特尽情展现。上衣胸口两边设有两个加大口袋，更增添了层次感。衣襟<UNK>了拉链，让穿脱更加的方便，轻轻<UNK>一点，显得更加时尚。”"]]}]
[{
"prompt": "类型#上衣*版型#宽松*风格#英伦*风格#复古*图案#格子*图案#复古*图案#线条*衣样式#外套*衣样式#西装*衣领型#翻领", "response": [["这件西装外套选用了经久不衰的格纹元素，通过色彩的明暗对比，展现出丰富的视觉层次，又缔造了英伦风的复古气息。法式的大翻领，延长颈部线条，彰显出女性帅气干练的特殊魅力。宽松舒适的版型完美掩藏了身材的小秘密，给身体自由活动空间。"]]}]
[{
"prompt": "类型#裙*版型#显瘦*材质#蕾丝*颜色#纯色*风格#知性*风格#高贵*风格#性感*图案#纯色*图案#蕾丝*裙型#背带裙*裙型#包臀裙*裙型#鱼尾裙*裙长#连衣裙*裙袖型#喇叭袖", "response": [["蕾丝喇叭袖上衣，搭配鱼尾包臀背带裙，整体造型给人甜美可人的感觉。偏爱蕾丝的浪漫柔情，流露别致女人味。喇叭袖的设计凸显别样浪漫，透露隐约小性感。两件套连衣裙，平添视觉层次感。鱼尾的设计修身显瘦，喇叭袖时尚减龄，纯色设计更加凸显女性知性高贵的气质。"]]}]
[{
"prompt": "类型#裙*风格#淑女*风格#清新*风格#性感*图案#碎花*图案#线条*裙型#大裙摆*裙下摆#荷叶边", "response": [["性感的挂脖领设计，展现出迷人的肩部线条，尽显女人的妩媚气息。清新的碎花点缀裙身，凸显出秀雅温柔的韵味，衬的人很是气质不俗。灵动的荷叶边装饰，让整件上衣多了一些柔美和俏皮。散开的大摆裙剪裁，修饰出身材的小缺陷，行走间尽显温婉的淑女气质。"]]}]
[{
"prompt": "类型#裙*材质#网纱*图案#蝴蝶结*裙下摆#层叠*裙长#半身裙*裙衣门襟#系带", "response": [["层叠网纱，仙气飘飘，却不会过于膨胀。腰间的蝴蝶结系带，恰到好处的增添了柔美感。膝盖以下，长度刚刚好的半身裙，比起“一览无遗魅力尽显”，专注于“完美隐藏”"]]}]
[{
"prompt": "类型#裙*版型#宽松*颜色#焦糖色*风格#简约*风格#ol*风格#职场*裙型#百褶*裙长#连衣裙*裙领型#翻领*裙款式#腰带*裙款式#衬衫式", "response": [["来自<UNK>自制的连衣裙采用今年大热的焦糖色，就像巧克力一样，甜蜜又不腻人。腰带的贴心设计，让宽松的版型也能拥有s曲线。上身简约的衬衫式翻领，衬托小v脸，带来一股职场ol风，加以百褶下摆的点缀，一起述说无尽温柔。"]]}]
[{
"prompt": "类型#裙*裙型#鱼尾裙*裙款式#收腰", "response": [["率性大方的裙身，加上经典百搭的收腰版型，轻松打造出了时尚大方感呢。更有着俏皮可爱的鱼尾裙摆以及靓丽打眼的鹅黄色裙身，尽显元气少女风范，减龄效果特别好还特别的有青春活力气息，适合各个年龄阶段的女生们穿着。"]]}]
[{
"prompt": "类型#上衣*颜色#红色*风格#青春*衣样式#外套*衣长#短款*衣款式#口袋", "response": [["这款外套对于个子矮小的妹纸来说就是福音了，短款穿在身上搭配起起来，立马就能变成大长腿，把整体身长比例拉长，呈现出黄金比例效果。鲜艳活泼的红色，穿在身上，视觉上给人呈现出青春的活力，元气满满的少女，还能衬托出肌肤的白皙，拥有一整天的好气色。大大的口袋，既可以作为装饰，出门携带东西也是非常的方便，还能增加整体的层次感。"]]}]
[{
"prompt": "类型#上衣*材质#牛仔布*颜色#浅蓝色*颜色#深蓝色*风格#休闲*风格#潮*衣样式#外套*衣款式#拼接*衣款式#口袋*衣款式#纽扣", "response": [["层叠网纱，仙气飘飘，却不会过于膨胀。腰间的蝴蝶结系带，恰到好处的增添了柔美感。膝盖以下，长度刚刚好的半身裙，比起“一览无遗魅力尽显”，专注于“完美隐藏"]]}]
[{
"prompt": "类型#裙*材质#网纱*图案#蝴蝶结*裙下摆#层叠*裙长#半身裙*裙衣门襟#系带", "response": [["BRAND牛仔外套，浅蓝色的衣身和深蓝色形成拼接的设计，充满了潮流的时尚感，翻折的领口造型，衬托在颈部肌肤，能修饰脸型。领口下有单排金属的纽扣门襟，开合很方便，很实用可以保暖。两侧有翻盖的口袋和斜插的口袋，在视觉上很有层次感。看起来很休闲。"]]}]
[{
"prompt": "类型#裙*版型#显瘦*材质#蕾丝*图案#蝴蝶结*图案#蕾丝*裙下摆#花边*裙下摆#压褶*裙长#半身裙*裙袖长#长袖*裙领型#立领*裙款式#拼接*裙款式#钉珠", "response": [["成熟韵味和优雅气质并存的时尚两件套。上衣立领系蝴蝶结造型，俏皮优雅。喇叭长袖拼接压褶蕾丝花边，气质减龄。高腰包臀半身裙，修身效果特别好，收腹<UNK>展示曼妙的身材曲线。两侧手工钉珠装饰，时髦立体，视觉拉长腿型，整体上身彰显成熟女人魅力。显瘦百搭。"]]}]
[{
"prompt": "类型#裙*材质#蕾丝*图案#刺绣*图案#蕾丝*裙衣门襟#拉链*裙款式#拉链*裙款式#吊带*裙款式#收腰", "response": [["蕾丝吊带显露出精致的锁骨，让颈部显得更加修长。腰部采用款腰围收腰的方式，小蛮腰更诱人。裙摆上大朵刺绣花朵，非常逼真，仿佛真正的花朵撒在裙子上摇曳生姿。背后贴心的珍珠扣，美观的同时又避免了生活中忘记拉拉链的尴尬情况，精致不失优雅。"]]}]
[{
"prompt": "类型#裙*版型#显瘦*图案#线条*裙下摆#花边*裙腰型#松紧腰*裙长#连衣裙", "response": [["连衣裙采用了松紧腰的设计，凸显出腰部纤细的线条，再加上过膝的长度，可以遮掩掉大腿上的小肉肉，更加显瘦，走路飘逸十足。采用了圆形领口的设计，修饰颈部线条，衣身上加了层次感分明的荷叶花边作为装饰，颇显甜美气质。"]]}]
Save the dataset as a data.jsonl file and save it locally!
import dataset
Click Dataset Management to enter the Create Dataset interface
Define the dataset name, select Text Dialog, Unsort, then create and import. Enter the interface for importing data.
Select label information, import locally, and the file format is jsonl. Then, click Upload jsonl file.
Select the dataset file data.jsonl.
Click OK, and then the data set can be uploaded to the Qianfan platform.
Preview these data, the data is relatively small! Easy fine-tuning and training.
data cleaning
Before training, you need to clean the data. Click Data Processing to enter the data cleaning interface.
Choose Dataset and Version, then choose Cleaned Configuration. Each cleaning has an example, so that everyone can check the effect!
data augmentation
Parallel to data cleaning is data augmentation, and we can also create data augmentation tasks.
Click Create Task to enter the data enhancement interface.
Then, select the dataset and augmentation method.
Publish dataset
After cleaning and enhancing the data, open the dataset management and publish the dataset we just created.
Large model tuning
Click Large Model Tuning to enter the Large Model Tuning interface and enable fine-tuning of the large model.
Click to create tuning task
Open payment.
Choose 7B local tuning! This one is the cheapest.
Click Next!
Then wait for the tuning task to be activated, and then click Create tuning task after waiting for activation.
Click Create Tuning Task.
Set commands and application scenarios, then click Create and Train.
Select the model, fine-tuning method, dataset, and click OK.
The task is created and training is in progress.
training.
Click on the details to view the training status.
RLHF training
Reward Model Training
Reward model training requires multiple rounds of dialogue datasets, so we first create a multi-round dialogue dataset.
Enter a dataset name, select With Sorting, and create label data.
I don't have a dialogue dataset, so I uploaded the official dataset. Then, continue data cleaning, enhancement, and publish the dataset. With the data set, the reward model training can be carried out.
Click on the ship training mission.
Enter the name, industry, and application scenario, and then click Create and Train.
Select the dataset you just uploaded, and then click OK.
Wait for a while to run to completion.
reinforcement learning training
Intensive training requires the use of query problem sets. So we need to upload a set of query questions. The data set is as follows:
[{
"prompt": "类型#上衣*材质#牛仔布*颜色#白色*风格#简约*图案#刺绣*衣样式#外套*衣款式#破洞"}]
[{
"prompt": "类型#裙*材质#针织*颜色#纯色*风格#复古*风格#文艺*风格#简约*图案#格子*图案#纯色*图案#复古*裙型#背带裙*裙长#连衣裙*裙领型#半高领"}]
[{
"prompt": "类型#上衣*风格#嘻哈*图案#卡通*图案#印花*图案#撞色*衣样式#卫衣*衣款式#连帽"}]
[{
"prompt": "类型#裤*风格#英伦*风格#简约"}]
[{
"prompt": "类型#裙*裙下摆#弧形*裙腰型#高腰*裙长#半身裙*裙款式#不规则*裙款式#收腰"}]
[{
"prompt": "类型#上衣*版型#宽松*版型#显瘦*图案#线条*衣样式#衬衫*衣袖型#泡泡袖*衣款式#抽绳"}]
[{
"prompt": "类型#裙*材质#蕾丝*风格#宫廷*图案#刺绣*图案#蕾丝*裙型#大裙摆*裙下摆#花边*裙袖型#泡泡袖"}]
[{
"prompt": "类型#裤*版型#显瘦*颜色#黑色*风格#简约*裤长#九分裤"}]
[{
"prompt": "类型#裙*版型#显瘦*风格#文艺*风格#简约*图案#印花*图案#撞色*裙下摆#压褶*裙长#连衣裙*裙领型#圆领"}]
[{
"prompt": "类型#裙*颜色#蓝色*风格#清新*图案#蝴蝶结"}]
[{
"prompt": "类型#裙*颜色#白色*风格#清新*图案#碎花*裙腰型#松紧腰*裙长#长裙*裙衣门襟#拉链*裙款式#拉链"}]
[{
"prompt": "类型#裤*材质#羊毛*裤长#九分裤*裤口#微喇裤"}]
[{
"prompt": "类型#上衣*风格#简约*衣门襟#拉链*衣款式#口袋*衣款式#拉链"}]
[{
"prompt": "类型#上衣*版型#宽松*风格#英伦*风格#复古*图案#格子*图案#复古*图案#线条*衣样式#外套*衣样式#西装*衣领型#翻领"}]
[{
"prompt": "类型#裙*版型#显瘦*材质#蕾丝*颜色#纯色*风格#知性*风格#高贵*风格#性感*图案#纯色*图案#蕾丝*裙型#背带裙*裙型#包臀裙*裙型#鱼尾裙*裙长#连衣裙*裙袖型#喇叭袖"}]
[{
"prompt": "类型#裙*风格#淑女*风格#清新*风格#性感*图案#碎花*图案#线条*裙型#大裙摆*裙下摆#荷叶边"}]
[{
"prompt": "类型#裙*材质#网纱*图案#蝴蝶结*裙下摆#层叠*裙长#半身裙*裙衣门襟#系带"}]
[{
"prompt": "类型#裙*版型#宽松*颜色#焦糖色*风格#简约*风格#ol*风格#职场*裙型#百褶*裙长#连衣裙*裙领型#翻领*裙款式#腰带*裙款式#衬衫式"}]
[{
"prompt": "类型#裙*裙型#鱼尾裙*裙款式#收腰"}]
[{
"prompt": "类型#上衣*颜色#红色*风格#青春*衣样式#外套*衣长#短款*衣款式#口袋"}]
[{
"prompt": "类型#上衣*材质#牛仔布*颜色#浅蓝色*颜色#深蓝色*风格#休闲*风格#潮*衣样式#外套*衣款式#拼接*衣款式#口袋*衣款式#纽扣"}]
[{
"prompt": "类型#裙*材质#网纱*图案#蝴蝶结*裙下摆#层叠*裙长#半身裙*裙衣门襟#系带"}]
[{
"prompt": "类型#裙*版型#显瘦*材质#蕾丝*图案#蝴蝶结*图案#蕾丝*裙下摆#花边*裙下摆#压褶*裙长#半身裙*裙袖长#长袖*裙领型#立领*裙款式#拼接*裙款式#钉珠"}]
[{
"prompt": "类型#裙*材质#蕾丝*图案#刺绣*图案#蕾丝*裙衣门襟#拉链*裙款式#拉链*裙款式#吊带*裙款式#收腰"}]
[{
"prompt": "类型#裙*版型#显瘦*图案#线条*裙下摆#花边*裙腰型#松紧腰*裙长#连衣裙"}]
[{
"prompt": "类型#上衣*材质#牛仔布*颜色#白色*风格#简约*图案#刺绣*衣样式#外套*衣款式#破洞"}]
[{
"prompt": "类型#裙*材质#针织*颜色#纯色*风格#复古*风格#文艺*风格#简约*图案#格子*图案#纯色*图案#复古*裙型#背带裙*裙长#连衣裙*裙领型#半高领"}]
[{
"prompt": "类型#上衣*风格#嘻哈*图案#卡通*图案#印花*图案#撞色*衣样式#卫衣*衣款式#连帽"}]
[{
"prompt": "类型#裤*风格#英伦*风格#简约"}]
[{
"prompt": "类型#裙*裙下摆#弧形*裙腰型#高腰*裙长#半身裙*裙款式#不规则*裙款式#收腰"}]
[{
"prompt": "类型#上衣*版型#宽松*版型#显瘦*图案#线条*衣样式#衬衫*衣袖型#泡泡袖*衣款式#抽绳"}]
[{
"prompt": "类型#裙*材质#蕾丝*风格#宫廷*图案#刺绣*图案#蕾丝*裙型#大裙摆*裙下摆#花边*裙袖型#泡泡袖"}]
[{
"prompt": "类型#裤*版型#显瘦*颜色#黑色*风格#简约*裤长#九分裤"}]
[{
"prompt": "类型#裙*版型#显瘦*风格#文艺*风格#简约*图案#印花*图案#撞色*裙下摆#压褶*裙长#连衣裙*裙领型#圆领"}]
[{
"prompt": "类型#裙*颜色#蓝色*风格#清新*图案#蝴蝶结"}]
[{
"prompt": "类型#裙*颜色#白色*风格#清新*图案#碎花*裙腰型#松紧腰*裙长#长裙*裙衣门襟#拉链*裙款式#拉链"}]
[{
"prompt": "类型#裤*材质#羊毛*裤长#九分裤*裤口#微喇裤"}]
[{
"prompt": "类型#上衣*风格#简约*衣门襟#拉链*衣款式#口袋*衣款式#拉链"}]
[{
"prompt": "类型#上衣*版型#宽松*风格#英伦*风格#复古*图案#格子*图案#复古*图案#线条*衣样式#外套*衣样式#西装*衣领型#翻领"}]
[{
"prompt": "类型#裙*版型#显瘦*材质#蕾丝*颜色#纯色*风格#知性*风格#高贵*风格#性感*图案#纯色*图案#蕾丝*裙型#背带裙*裙型#包臀裙*裙型#鱼尾裙*裙长#连衣裙*裙袖型#喇叭袖"}]
[{
"prompt": "类型#裙*风格#淑女*风格#清新*风格#性感*图案#碎花*图案#线条*裙型#大裙摆*裙下摆#荷叶边"}]
[{
"prompt": "类型#裙*材质#网纱*图案#蝴蝶结*裙下摆#层叠*裙长#半身裙*裙衣门襟#系带"}]
[{
"prompt": "类型#裙*版型#宽松*颜色#焦糖色*风格#简约*风格#ol*风格#职场*裙型#百褶*裙长#连衣裙*裙领型#翻领*裙款式#腰带*裙款式#衬衫式"}]
[{
"prompt": "类型#裙*裙型#鱼尾裙*裙款式#收腰"}]
[{
"prompt": "类型#上衣*颜色#红色*风格#青春*衣样式#外套*衣长#短款*衣款式#口袋"}]
[{
"prompt": "类型#上衣*材质#牛仔布*颜色#浅蓝色*颜色#深蓝色*风格#休闲*风格#潮*衣样式#外套*衣款式#拼接*衣款式#口袋*衣款式#纽扣"}]
[{
"prompt": "类型#裙*材质#网纱*图案#蝴蝶结*裙下摆#层叠*裙长#半身裙*裙衣门襟#系带"}]
[{
"prompt": "类型#裙*版型#显瘦*材质#蕾丝*图案#蝴蝶结*图案#蕾丝*裙下摆#花边*裙下摆#压褶*裙长#半身裙*裙袖长#长袖*裙领型#立领*裙款式#拼接*裙款式#钉珠"}]
[{
"prompt": "类型#裙*材质#蕾丝*图案#刺绣*图案#蕾丝*裙衣门襟#拉链*裙款式#拉链*裙款式#吊带*裙款式#收腰"}]
[{
"prompt": "类型#裙*版型#显瘦*图案#线条*裙下摆#花边*裙腰型#松紧腰*裙长#连衣裙"}]
Click Dataset Management, enter the name of the dataset, and set the label type to query question set. Then click Create and Import
to import locally, the format is jsonl, and then upload the file. After uploading, do data cleaning, enhancement, and then publish the data.
Activate reinforcement learning training, create training tasks, and activate the business.
Select Reinforcement Learning, and then the next step.
continue to the next step
congratulations! Successful activation!
Then, go ahead and create a training task!
Enter the name, industry, application scenario, and click create and train!
Select the dataset you just created, select the fine-tuned model and the reward model. Then, click OK.
Wait for the run to complete! ! !
Large model management
After waiting for the model to run, we can create the model on the model management page. Large model management includes model management, model evaluation, and model compression.
model management
This operation is similar to the previous operation, fill in the model name, model version, select the training task, and run the name. Then click OK.
Then you can see the model you just created.
Click on the details to further compress, evaluate and deploy the model! Click Details to see detailed information such as the basic information of the model, interface information, and instructions for use.
model evaluation
Next, click Model Evaluation to evaluate the trained model! ! !
Enter a name, select a model, select a dataset for evaluation, and click OK.
Wait for the run to complete! !
Then you can see the evaluation results! The model scored poorly due to limited data and time to train!
Large model service
service management
After the evaluation is complete, the model can be deployed according to the situation of the model. Click Service Management. Create a service, you can see the following interface:
Select model, set version and service name, interface address. Then click Next.
Choose time and computing power, then apply for release! A computing power unit is 250 a day! ! ! Apply for an online test! I will not spend this money, everyone can try it if they need it!
Summarize
The above is my tutorial summarizing the large model of Qianfan. The entire platform has no complicated functions, and students who have a certain knowledge reserve of AI can easily get started. I wish you all can train your own large model as soon as possible!
