This blog post focuses on new features and improvements. For a comprehensive list, including bug fixes, please see Continued note.
We are introducing pre-built, ready-to-use templates that simplify the app creation process for various use cases. Each template comes with a range of resources, such as datasets, models, workflows, and modules, that allow you to quickly start your app creation process.
When you choose a template to build an app, the configurations and resources available in the template will be already applied to your new application. You can use pre-built components to quickly implement AI in your specific use case.
Recent releases have added new templates such as sentiment analysis and text moderation, along with existing templates. Let's see the details of each template:
Here are some of the workflows available in the Text Moderation template for different use cases:
1. A Multilingual Moderated Grading Workflow: This is a wrap around workflow A Model of Multilingual Text Moderation which classifies text as toxic, offensive, obscene, identify_hate and severe_toxic.
2. English text moderation classifier workflow: This is a wrap around workflow English text moderated model which classifies English text as toxic, insulting, obscene, identify_hate and severe_toxic.
3. Text-Moderation-mistral-7b: Uses a workflow. Mistral-7b Model with specific prompt template for text moderation that identifies and Filter out any hate speech, violent language and explicit content and respond with 'inappropriate' if such content is present and 'appropriate' otherwise.
4. text-moderate-misinformation-dbrx: Uses a workflow. DBRX Model for moderation of misinformation with a specific prompt template that identifies and Filter out misinformation or unsubstantiated claims, especially related to health, science, or news events. And reply with 'possible misinformation'.
5. LlamaGuard-Prompt-Moderate: Uses a workflow. Llama Guard Model With a specific prompt template to protect LLMs input prompts.
Here are some workflows from the template.
1. Tweet Sentiment Analysis Workflow: It uses workflow. twitter-roberta-base-sentiment-latest Model tRained over ~124M tweets, and optimized for sentiment analysis. With the TweetEval benchmark.
2. Multilingual Sentiment Analysis-DistilBERT Workflow: It uses workflow. Distillbert-base-multilingual-cased-emotions The model is suitable for sentiment analysis in multiple languages.
3. Financial Sentiment Analysis – Finbert Workflow: It uses workflow. Finberta BERT-based model fine-tuned on financial text for high-accuracy sentiment analysis in the finance domain.
4. Emotional-Analysis-Mistral-7b Workflow: Uses a workflow. Mistral-7b A specific prompt template for sentiment analysis models and predicts whether a sentence/paragraph is positive/negative or neutral sentiment.
Contains the template. of Clarify Chatbot module Which lets you chat with several major language models with a single UI interface.
Here are a few models and workflows that include the Image Moderation template.
1. NSFW recognition model Predicts the likelihood that an image contains provocative or sexually explicit nudity. This is a great solution for those looking to automatically moderate or filter nudity from their platform. This is limited to specific use cases of nudity.
2. NSFW-Recognition workflow: This workflow is wrapped. NSFW recognition classifier which classifies images as nudity, sexually explicit images.
3. Hate Symbol Detection Workflow: This workflow has a wrap around Hate Symbol Detector which detects the ADL's recognized hate symbol.
This app template discusses several aspects of content creation like email writing, blog writing, question answering, storytelling, social media content and comes with several ready-to-use workflows. Content creation.
This template explores 3 basic approaches to summarizing, starting with Novice and ending with Expert.
1. Summarizing Sentences and Paragraphs: It helps to summarize a few paragraphs and you want to summarize them. You can easily submit text for summarization via the following workflow, and it will return a condensed version:
2. Page Level Summarization: This method solves the challenge of summarizing text spanning multiple pages.
3. Summarize the whole book: This method can help to summarize the whole book with best representation vector method.
The app template includes pre-built RAG agents, leveraging various LLM models and optimized through various prompt engineering techniques, to name a few.
1. Rag-agent-gpt4-turbo-naive: It uses the RAG agent. GPT-4 Turbo LLM model with a simple prompt for direct integration.
2. Rag-agent-claude2-1-CoT-few-shot: This RAG agent is used with the Claude-2.1 LLM model. Indicating CoT For better reasoning and performance.