# Predibase

> Predibase was a San Francisco-based AI infrastructure company (founded 2020, acquired by Rubrik in June 2025) that pioneered efficient LLM fine-tuning and serving at scale. Built by the creators of Uber AI's Ludwig and Horovod frameworks, Predibase made it easy for enterprises to fine-tune and deploy open-source LLMs using LoRA adapters — often outperforming GPT-4 on domain-specific tasks for under $8 of compute. Its open-source LoRAX inference server enabled serving thousands of fine-tuned models from a single GPU, dramatically cutting costs. After raising $28M from Greylock and Felicis, Predibase was acquired by cybersecurity firm Rubrik for over $100M to accelerate agentic AI adoption.

- **Founded:** 2020
- **Headquarters:** San Francisco, CA, USA
- **Founders:** Piero Molino (Co-Founder & original CEO), Travis Addair (Co-Founder & CTO), Devvret Rishi (Co-Founder & CEO (succeeded Molino)), Chris Ré (Co-Founder & Academic Advisor)
- **Team size:** ~25–34 employees at time of acquisition (June 2025)
- **Products:** Predibase Fine-Tuning Platform, LoRAX (LoRA eXchange), Predibase Inference Engine, Reinforcement Fine-Tuning (RFT) Platform, LoRA Land
- **Notable:** Raised $28.4M total from top-tier investors Greylock and Felicis Ventures, Acquired by Rubrik for over $100M in June 2025 — one of the first major acquisitions of an LLM fine-tuning infrastructure company, Built LoRAX, open-source multi-LoRA inference server with 3,700+ GitHub stars

## Products & services

- **Predibase Fine-Tuning Platform** — Managed, serverless platform for supervised fine-tuning (SFT) of open-source LLMs using LoRA adapters. Supports multi-GPU training, large datasets, function-calling fine-tuning, and incremental/continuous model updates.
- **LoRAX (LoRA eXchange)** — Open-source multi-LoRA inference server that serves thousands of fine-tuned LLMs from a single GPU using dynamic adapter loading and heterogeneous continuous batching. 3,700+ GitHub stars.
- **Predibase Inference Engine** — Next-generation proprietary inference stack combining Turbo LoRA, LoRAX, and FP8 quantization. Delivers 3–4x throughput vs. base models and 50%+ cost reduction. Launched October 2024.
- **Reinforcement Fine-Tuning (RFT) Platform** — First fully managed, serverless RFT SDK using GRPO (Group Relative Policy Optimization). Enables reasoning model fine-tuning with as few as 10–100 labeled examples. Launched February 2025.
- **LoRA Land** — A curated collection of 25+ fine-tuned Mistral-7B models that outperform GPT-4 on domain-specific tasks, fine-tuned for under $8 each on average, all served from a single NVIDIA A100 GPU. Launched February 2024.
- **Predibase VPC** — Virtual Private Cloud deployment option for data-sensitive enterprises who need isolated GPU infrastructure for fine-tuning and inference.

## Achievements

- Raised $28.4M total from top-tier investors Greylock and Felicis Ventures
- Acquired by Rubrik for over $100M in June 2025 — one of the first major acquisitions of an LLM fine-tuning infrastructure company
- Built LoRAX, open-source multi-LoRA inference server with 3,700+ GitHub stars
- LoRA Land demonstrated fine-tuned 7B models beating GPT-4 on domain tasks for under $8 of compute (Feb 2024)
- First fully managed Reinforcement Fine-Tuning (RFT) platform for open-source LLMs (Feb 2025)
- First provider offering private on-demand Qwen 3 endpoints on AWS (May 2025)
- Helped Marsh McLennan save over 1 million team hours in year one using agentic AI
- Over 10,000 SLMs fine-tuned on the platform; 700+ LLM fine-tuning experiments published in Fine-Tuning Index
- Founders co-created Ludwig (9K+ GitHub stars) and Horovod (13K+ GitHub stars) at Uber AI
- Predibase Inference Engine delivered 3–4x throughput improvement and 50%+ cost savings (Oct 2024)

## Latest updates

- **2025-06** — Acquired by Rubrik (NYSE: RBRK) for over $100M. Predibase technology powers Rubrik Agent Cloud for agentic AI governance in enterprise data security.
- **2025-05** — Became first and only provider offering private, on-demand Qwen 3 model endpoints on AWS infrastructure.
- **2025-02** — Launched Reinforcement Fine-Tuning (RFT) platform — industry's first fully managed, serverless RFT SDK using GRPO, enabling reasoning model training with as few as 10 labeled examples.
- **2024-10** — Launched Predibase Inference Engine combining Turbo LoRA, LoRAX, and FP8 quantization for 3–4x throughput and 50%+ cost reduction.
- **2024-02** — Launched LoRA Land — 25+ fine-tuned Mistral-7B models that outperform GPT-4 on domain-specific tasks, served from a single GPU for under $8 each.

## Links

- Website: https://predibase.com
- LinkedIn: https://www.linkedin.com/company/predibase
- Twitter/X: https://x.com/predibase
- GitHub: https://github.com/predibase

---

Profile page: https://yespress.io/predibase
Published by YesPress — https://yespress.io
Last updated: 2026-04-02