OVH just announced its first LLM, Olala mini. It has 7B parameters with only 1B active at a time, combining attention with Mamba-3. They claim it runs on an 8GB GPU, or even on a CPU at 50 tokens/sec. They’re planning to release it on Hugging Face under Apache 2.0.
OVH just announced its first LLM, Olala mini. It has 7B parameters with only 1B active at a time, combining attention with Mamba-3. They claim it runs on an 8GB GPU, or even on a CPU at 50 tokens/sec. They’re planning to release it on Hugging Face under Apache 2.0.