RAG in 4 lines of code

This blog post focuses on new features and improvements. For a comprehensive list, including bug fixes, please see Continued note.

RAG in 4 lines of code

RAG is an architecture that provides LLMs with the most relevant and contextually significant data when answering questions. You can use it for applications such as advanced question answering systems, information retrieval systems, chatting with your data, and more.

We have integrated the new RAG-Prompter operator model. Now you can use RAG-Prompter, an agent system operator in the Python SDK, to perform RAG tasks in just 4 lines of code.

Check the following. video Which walks you through the step-by-step process of building a RAG system in 4 lines of code.

Integrated Clarify into DSPy.

• DSP A framework for solving advanced tasks with language and retrieval models. It combines techniques for sharpening and fine-tuning language models.
  
  This integration, now a part of the recently released DSPy version 2.1.7, helps you use Clarifai's LLM models and use your Clarifai apps as a vector search engine within DSPy. Is. Clarifai is the only provider that enables users to use multiple LLM models. You can get started on how to use DSPy with Clarify. Here.

Introduced additional model version training

• Now you can update existing models with new data without retraining them from scratch. After training a model version, a checkpoint file is automatically saved. You can start additional training from an already trained version checkpoint. Alternatively, you provide the URL of the checkpoint file from a supported 3rd party toolkit such as HuggingFace or MMCV.

Introduced the ability to add input via cloud storage URLs

• You can now provide URLs with required access credentials from cloud storage platforms such as S3, GCP, and Azure. This functionality makes it easy to add inputs to our platform, offering a more efficient alternative to the traditional method of using PostInputs for individual inputs.

Improved the evaluation process for detector models

• Enhanced the matrix by introducing additional fields, namely “Total Prediction,” “True Positive,” “False Negative,” and “False Positive.” These additional metrics provide a more comprehensive and detailed assessment of detector performance.
• Previously, a multiselector was used to select an intersection at a union (IoU). We replaced this ambiguous selection with a radio button format, emphasizing a single, mutually exclusive choice for IoU selection.
• We have also made other minor UI/UX improvements to ensure consistency in the evaluation process of classification models.

LLM has improved fine-tuning.

• Added support for CSV upload for seamless data integration.
• Added more training templates to fine-tune the process to suit diverse use cases.
• Added advanced configuration options, including quantization parameters via GPTQ, which further empowers users to optimize models with greater accuracy and efficiency.
  

Improved the ModelViewer's version table

• Cross-app evaluation is now supported in the Model Versions tab for a more integrated experience with the leaderboard.
  
• Users and contributors with access permissions can also select datasets or dataset versions from org apps, ensuring comprehensive evaluation in different contexts.
• This improvement allows users to view training and evaluation data across different model versions in a centralized location, enhancing the overall version tracking experience.

Improved model annotations and related asset management

• Previously, when a model annotation was deleted, the associated model assets were not affected. If you now delete a model annotation, a simultaneous action will mark the corresponding model assets as deleted. This ensures that the deletion process is comprehensive, avoiding delays or orphaned assets.

Published several new, ground-breaking models.

• Published Phi-2, a Clarifai-hosted, 2.7 billion parameter Large Language Model (LLM), achieving cutting-edge performance in QA, chat, and code tasks. It focuses on high-quality training data and has demonstrated improved behavior in toxicity and bias.
• Wrapped up Depgram Nova-2. It sets a new benchmark in speech-to-text with a 30% lower error rate and unmatched speed, making it the best choice in automatic speech recognition.
  
• Wrapped up Synopsis of Diagram Audio. It offers efficient and accurate summarization of audio content, automating call notes, meeting summaries, and podcast previews with superior transcription capabilities.
• Wrapped up text-embedding-3-largeA high performance, flexible text embedding model with up to 3072 dimensions, outperforming its predecessor.
  
• Wrapped up text-embedding-3-smallA highly efficient, flexible model with better performance on various natural language processing tasks than its predecessor, Text-Embedding-ADA-002.
• Wrapped up CodeLlama-70b-instructiona state-of-the-art AI model that specializes in code generation and instruction-based understanding of natural language.
• Wrapped up CodeLlama-70b-PythonA state-of-the-art AI model that specializes in generating and understanding Python code, excels in accuracy and efficiency.

Improved the mobile version of the onboarding flow

• Updated the “Build an App” guided tour model for mobile platforms.
• Made other improvements such as updating the “Add a model” model and the “Find a pre-trained model” model for mobile platforms.

Added the ability to minify existing image mask annotations on the input viewer.

• You can view your image mask annotations uploaded via the API.
• You can delete the entire image mask annotation on the input.
• You can view the mask annotation items displayed on the Input Viewer sidebar.
  

Made minor enhancements to the Workflow Builder UI.

• Corrected alignment differences in some left side models to ensure uniform left alignment.
  
• Introduced an X or Close/Cancel button for better user interaction and clarity.
• It ensured that users can easily straighten the line connecting two nodes.
  

Added the ability to copy an app to an organization.

• Before that, I Copy/Duplicate app In the modal, the dropdown to select users lacked an option for organizations. You can now directly select an organization from the drop-down list of possible destinations when copying or duplicating an app.
  

Improved the search behavior in the use_cases field

• Before that, the use_cases The field within the ListModels feature was configured as an AND search, unlike other fields input_fields And output_fields. We did better use_cases Attribute to work with OR logic, just like other fields. This adjustment broadens the scope of the search results, accommodating scenarios where models may apply to diverse use cases.
  

Changed the thumbnails for the resource list to use smaller versions of the cover images.

• Previously, thumbnails for resource listings used larger versions of cover images. We changed them to use smaller versions — just like for other resources such as apps, models, workflows, modules, and datasets. We have also changed the left sidebars.

Implemented a modification to provide a more user-friendly experience for non-logged-in users interacting with text-to-image models.

• Clicking the “Generate” button now triggers the login/signup popup modal. It guides users who are not currently logged in through the necessary authentication steps to ensure a smooth transition to using the model's functionality.

Fixed an issue where a user could be added to the same organization multiple times.

• We have implemented safeguards against unintentional duplication of users within an organization. Previously, if a user clicked the “Accept” button multiple times on an organization's invitation page, they could be registered to the same organization in vain. Consequently, the user interface displayed multiple instances of the same organization.

Improved the module installation process

• Modal has been optimized to use app IDs, removing the dependency on deprecated app names. Previously, the popup modal for installing modules into an app retained the use of deprecated app names.

Improved the compatibility of the GitHub link on the module page

• Previously, a small GitHub button was at the top of any module's overview page. We moved it to the right, aligning it with other metadata such as the description, thus improving its clarity as a clickable link.