Open models, built from the ground up.
Our LLMs, SLMs and diffusion models are used by more than 60,000 developers, startups and enterprises. Over 20 are open-sourced, engineered and trained from scratch to run efficiently on client devices and at the edge.
Four families, seventeen models.
GenZ for general instruction following, Code Millennials for code generation, SQL Millennials for text-to-SQL, and Boomer for small models pre-trained on synthetic data. Every one is published openly.
GenZ 70B
An advanced LLM fine-tuned on the foundation of Meta’s open-source Llama V2 70B.
GenZ Vision Assistant 7B
A multimodal model fine-tuned to understand text and visual input together.
GenZ 13B V2 (ggml)
Instruction fine-tuned, 4K input length, on top of pretrained LLaMa2.
Code Millennials 34B
State-of-the-art open code generation, outperforming existing models across sizes.
Code Millennials 13B
State-of-the-art open code generation, outperforming existing models across sizes.
Code Millennials 8B
State-of-the-art open code generation, outperforming existing models across sizes.
Code Millennials 3B
State-of-the-art open code generation, outperforming existing models across sizes.
Code Millennials 1B
State-of-the-art open code generation, outperforming existing models across sizes.
SQL Millennials 13B
Built on CodeLLaMa 13B, fine-tuned on 100k SQL query generation instructions.
SQL Millennials 7B
Built on Mistral 7B, fine-tuned on 100k SQL query generation instructions.
Boomer 4B
A 3.51B-parameter model pre-trained on custom synthetic, textbook-style data.
Boomer BitNet 634M
A 634M BitNet model pre-trained from scratch on 5B tokens of synthetic data.
Need a domain-tuned model on your own hardware?
Bud Model Foundry trains and post-trains open models on infrastructure you already own. A solutions architect can scope it.