This weblog submit focuses on new options and enhancements. For a complete record, together with bug fixes, please see the launch notes.
We’re introducing pre-built, ready-to-use templates that simplifies the app creation course of for numerous usecases. Every template comes with a variety of assets, akin to datasets, fashions, workflows, and modules, permitting you to shortly get began together with your app creation course of.
If you select a template to create an app, the configurations and assets obtainable within the template shall be preemptively utilized to your new software. You should utilize the pre-built parts to shortly apply AI to your particular use case.
Together with the prevailing templates within the current launch, added new templates akin to Sentiment Evaluation and Textual content Moderation. Let’s take a look at the small print of every template:
Listed here are a number of the workflows obtainable within the Textual content Moderation template for numerous usecases:
1. Multilingual moderation Classifier Workflow: This workflow is wrapped across the Multilingual textual content moderation mannequin that classifies textual content as poisonous, insult, obscene, identity_hate and severe_toxic.
2. English textual content moderation Classifier Workflow: This workflow is wrapped across the English textual content moderation mannequin that classifies english textual content as poisonous, insult, obscene, identity_hate and severe_toxic.
3. Textual content-moderation-mistral-7b: Workflow makes use of Mistral-7b Mannequin with specified immediate template for textual content moderation that determine and filter out any hate speech, violent language and express content material and Reply with ‘Inappropriate’ if such content material is current and ‘Acceptable’ in any other case.
4. Textual content-moderation-misinformation-dbrx: Workflow makes use of DBRX Mannequin with specified immediate template for Misinformation moderation that determine and filter out misinformation or unsubstantiated claims, particularly associated to well being, science, or information occasions and reply with ‘Potential Misinformation’ if the content material appears questionable or ‘Doubtless Dependable’ if the knowledge seems to be credible.
5. LlamaGuard-prompt-moderation: Workflow makes use of Llama Guard Mannequin with specified immediate template for security of LLMs enter prompts. 
Listed here are just a few workflows from the template.
1. Tweet sentiment Evaluation Workflow: This workflow makes use of twitter-roberta-base-sentiment-latest mannequin trained on ~124M tweets, and finetuned for sentiment evaluation with the TweetEval benchmark.
2. Multilingual Sentiment Evaluation-DistilBERT Workflow: This workflow makes use of distilbert-base-multilingual-cased-sentiments mannequin, finetuned for sentiment evaluation in a number of languages.
3. Monetary sentiments Evaluation-FinBERT Workflow: This workflow makes use of finbert, A BERT-based mannequin fine-tuned on monetary textual content for high-accuracy sentiment evaluation within the finance area.
4. Sentimental-analysis-mistral-7b Workflow: Workflow makes use of Mistral-7b Mannequin with a specified immediate template for sentimental evaluation and predicts if the sentence/ pargraph is constructive/damaging or impartial sentiment.
The template consists of Clarifai’s Chatbot Module that permits you to chat with a number of Giant Language Fashions with a single UI interface.
Listed here are just a few fashions and workflows that the Picture Moderation Template consists of.
1. NSFW Recognition mannequin predicts the chance that a picture accommodates suggestive or sexually express nudity. It is an excellent resolution for anybody making an attempt to reasonable or filter nudity from their platform routinely. It’s restricted to nudity-specific use circumstances.
2. NSFW-recognition workflow: This workflow is wrapped round NSFW Recognition Classifier that classifies the photographs as nudity, sexually-explicit imagery.
3. Hate-symbol-detection workflow: This workflow is wrapped round Hate symbol-detector that detects the ADL-recognized hate image.
This App Template discusses a number of content material technology use circumstances akin to E mail writing, Weblog writing, Query Answering, Storytelling, Social media content material and comes with a number of ready-to-use workflows for content material creation.
This template discover 3 major strategies for summarization that begin with Novice and find yourself with Skilled.
1. Sentence and Paragraph Summarization: This helps to summarise just a few paragraphs and need to one-off summarize. You possibly can merely submit the textual content for summarization by the next workflow, and it’ll return a condensed model:
2. Web page-level Summarization: This technique addresses the problem of summarizing texts spanning a number of pages.
3. Summarize a whole ebook: This technique can assist to summarise your entire ebook with the Greatest Illustration Vectors Technique.
The App template contains pre-built RAG brokers, leveraging completely different LLM fashions and optimized by numerous immediate engineering strategies, to call just a few.
1. Rag-agent-gpt4-turbo-naive: This RAG Agent makes use of GPT-4 Turbo LLM mannequin with a easy immediate for simple integration.
2. Rag-agent-claude2-1-CoT-few-shot: This RAG Agent makes use of the Claude-2.1 LLM mannequin with CoT prompting for enhanced reasoning and efficiency.