OVH launches first in-house LLM family: Olala mini 7B with 1B active params, runs on 8GB GPUs
jedisct1 · x · 2026-11-17
French cloud provider OVH has announced Olala, its first LLM family, debuting with Olala mini 7A1B — a 7B-parameter model with 1B active params spread across 256 experts per layer, using hybrid attention with Differential-TPA and Mamba-3 in BF16.
Only a fraction of the model activates per request, so it runs on any GPU with 8GB+ VRAM: 160 tokens/s on an RTX A4500 and 50 tokens/s on a plain CPU, with a 32K-token context window. A promotional comic claims training costs dropped 20x since 2023 and that European search engine Qwart has adopted it. It will be open-sourced on Hugging Face.
Related event: OVH Releases First In-House Open-Source MoE Model Olala mini(2 posts)→
More from Infra
- Engineer runs Gemma 3 270M entirely inside an office laser printer at 60s per token — osanseviero · 2026-11-17
- Older GPUs abandoned by new inference engines: are they still worth keeping? — pipecanon · 2026-11-17
- Blogger argues total compute controls create blind spots and accelerate adversaries — TinfoilTricorn · 2026-11-17
- How rollout staleness hurts on-policy distillation, and how to build a faster pipeline — burkov · 2026-11-17
- SK hynix's Nvidia revenue doubled to $24B, but US customer concentration is the real risk — tengyanAI · 2026-11-17
- macOS TCP Stack Breaks After Exactly 49.7 Days of Uptime Due to 32-bit Timestamp Overflow — DanielLockyer · 2026-11-17