Open Source h2oGPT with Retrieval Augmented Generation (RAG), Web Search, and Agents
Download as PPTX, PDF0 likes657 views
The document discusses various AI models and tools developed by H2O.ai, including their focus on enterprise applications that emphasize security, scalability, and customization. Key offerings include integrations with popular LLMs, support for RAG frameworks, and a fully containerized, customizable platform for enterprise use. It highlights benchmarks for open-source LLMs and emphasizes ensuring data ownership and adaptability across different cloud environments.
Open Source h2oGPT with Retrieval Augmented Generation (RAG), Web Search, and Agents
- 1. H2O.ai Confidential Arno Candel Chief Technology Officer, H2O.ai Jon McKinney Director of Research, H2O.ai S P E A K E R S
- 3. v H2O.ai Confidential RAG PrivateGPT, Quiver, h2oGPT Agency AutoGPT, AutoGen, BabyAGI, ChatDev, GPT-Engineer Inference + Data vLLM, Semantic Kernel, DSPy, Langchain, LlamaIndex, MemGPT H2O Data Studio, H2O LLM Eval (Pre-) Training OpenAI, Meta, Google, Mistral Torch, DeepSpeed Fine-Tuning LLM Studio, Transformers LORA, RLHF, DPO Hardware AI Clouds NVIDIA GPUs AI Stack
- 5. H2O.ai Confidential Application: ❏ Multi-Chat ❏ RAG w/HYDE ❏ Query w/Web ❏ Map & Map Reduce ❏ Python & Search Agents Focus: ❏ KGM Models ❏ Parallelism ❏ Instruct LLMs ❏ Prompt Engineering ❏ Testing new LLMs Goal: ❏ OSS RAG with API ❏ Backend for Enterprise h2oGPT
- 9. H2O.ai Confidential What LLM saw Where was the answer?
- 14. H2O.ai Confidential KGM Image-OCR Models
- 15. v H2O.ai Confidential Challenges Context Size: ❏ 100k - Claude2 ❏ 128k - OpenAI GPT-4 ❏ 32k - LongAlpaca 70B LLaMa-2 ❏ 128k - Yarn - Mistral 7B ❏ 192k - Baichuan2 Other Challenges: ❏ Vague or Meta questions ❏ Critique, Verification ❏ Lexical + Semantic ❏ Retrieval vs. Summarization and Extraction ❏ Time-Aware RAG ❏ Multi-Modal Agency Meta Queries Retrieval Understanding
- 17. H2O.ai Confidential Enterprise h2oGPTe h2oGPTe = h2oGPT + focus on Enterprise Requirements < Security, Scalability, GUI, Collaboration, APIs, Airgapped, Multi-Cloud, HA, K8s /> AI includes: HyDE, ReRanker, Self-reflection, Prompt Engineering, etc.
- 18. H2O.ai Confidential Enterprise h2oGPTe Fully customizable APIs!
- 19. H2O.ai Confidential h2oGPTe RAG Benchmarks Key Take-Aways for h2oGPTe: ● open-source LLMs are good enough ● built on top of h2oGPT for latest AI ● designed for airgapped environments ● runs on any cloud/VPC, K8s ● Python/JS client API + GUI ● fully containerized and scalable ● customizable, any LLM, any language ● guard rails, agents, tools on road map ● own your data and models! https://h2o.ai/platform/enterprise-h2ogpt/ https://pypi.org/project/h2ogpte/ < DEMO BOOTH! See live demos! />