Aleph-Alpha/Kolibri-1-BF16
archivedmodel
Kolibri is Aleph Alpha's mixture-of-experts (MoE) reasoning model, with a focus on German and English. Mixture-of-experts, 78B total (384 experts, 6 per token). Context: 256K tokens (config). License: apache-2.0.
- Org
- Aleph-Alpha
- Released
- 2026-10-02
- Status
- archived
- License
- apache-2.0 (permissive)
- Parameters
- 78.1B
- Architecture
- MoE · 384 experts, 6/token · Kolibri1ForCausalLM
- Context
- 262,144 tokens
- Task
- text-generation · text
- Signal
- 63 HF likes · 2931 downloads
- Archived
- 2026-10-21 (no longer listed by the collector; stats as last seen)
Runs on
real quant files; need = weights + FP16 KV cache for 8K tokens + 1GB. A sizing calculation, not tested.
- Q4 ~51.6 GB: Mac 96GB, CPU only, 64GB RAM (slow)
- Q8 ~84.9 GB: 96GB GPU (RTX Pro 6000 Blackwell), 1× DGX Spark (128GB), Strix Halo 128GB unified (~96GB usable), Mac 128GB, Mac 192GB, CPU only, 128GB RAM (slow)
- FP16 ~158.1 GB: 2× DGX Spark (256GB), Mac 256GB, Mac 512GB, CPU only, 256GB RAM (slow)
Links
First published here 2026-10-02 · updated 2026-10-21 · markdown · JSON