Arize AI are the creators of Arize, a Machine Learning Observability platform that helps ML practitioners successfully take models from research to production with ease. They also created and maintain the open source project Arize Phoenix, that helps users evaluate, troubleshoot, and fine tune LLM, CV, and NLP models in a notebook.
Arize AI and Zilliz
Together Arize AI and Zilliz help users better understand and fine tune their LLM, CV, and NLP models to increase confidence in their embeddings and the retrieval augmented generation (RAG) and similarity search systems that they build with vector databases like Zilliz Cloud and Milvus.
RAG Evals: Statistical Analysis of Retrieval Strategies
In this video, Jason Lopatecki CEO and Co-Founder and Sally-Ann DeLucia ML Solutions Engineer at Arize AI delve into the 5 pillars of LLM Observability: evaluation, traces & spans, prompt engineering, search & retrieval, and fine-tuning. These pillars explore LLM output evaluations, context retrieval enhancement, and insights on benchmarking and analyzing retrieval systems for LLMs and RAGs.
Building And Troubleshooting An Advanced LLM Query Engine
In this on-demand session, go step-by-step on creating a robust query engine using the combined power of Arize Phoenix, LlamaIndex, LangChain, GPT 3.5 Turbo, NLTK, and Milvus. You’ll get an overview of LLM orchestration, an introduction to vector databases and an explanation how search and retrieval works and why it is needed.
Embeddings: Discover the Key To Building AI Applications That Scale with Zilliz
This conference talk focused on using embeddings for scalable generative AI applications. It discusses how the CVP framework can be used to fix a lot of the existing issues around hallucination and the lack of domain knowledge that we see in generative AI models. They see a demo of OSS chat, a manifestation of the CVP framework.
Extending the Context Window of LLaMA Models Paper Reading
Listen to AI & ML experts discuss a research paper on Position Interpolation (PI), a method extending context window sizes of LLaMA models up to 32,768 positions with minimal fine-tuning. You’ll learn how attention scores work in order to understand what positional embeddings are really there for.