Changelog
< All posts
January 15, 2025
ModernBert, more DSPy support and Qwen 2.5
New models and features have landed in Relevance!
1. ModernBert for embeddings in Advanced RAG
2. More DSPy support
3. Added Qwen as an agent model
DETAILS
- ModernBERT for Advanced RAG - We have just released ModernBert (released by the answer.ai team) to our AdvancedRAG. We are utilising ModernBert Embed by Nomic-AI which is an embedding model trained from ModernBERT-base, bringing the new advances of ModernBERT to embeddings! More details on the model are available here: https://huggingface.co/nomic-ai/modernbert-embed-base
- Expanded DSPy Support - We have added more support & integration for the DSPy framework to the Relevance platform with tool steps like - DSPy get training data, DSPy train and DSPy run to train DSPy system to generate responses based on gold set answers.
- Qwen 2.5 Joins Relevance Agents - We're expanding our tool-calling model support with the addition of Qwen 2.5 to Relevance Agents. Qwen has demonstrated exceptional performance in coding and tool calling capabilities, achieving benchmark results comparable to GPT-4o. This addition further strengthens our suite of available models for complex agents.
Other improvements
A smoll one but a good one! --> The entry point for viewing agent queues has been restyled!
At a glance you can now see how many active queues your agent has and how many upcoming items are going to be processed in those active queues. For better user experience, the button is now also hidden if the agent has no queues (active or paused). The scheduled triggers queue is now only created when the agent enables that ability instead of by default on agent creation.
