Kolibri-1 (78B-A3.5B)
European Apache-2.0 MoE: 78B total, 3.46B active, 262K native context (1M by extrapolation). Reasoning effort low/high or none.
Recommended quants
- Q4_K_M (47.5 GB)
Hob-forge/Kolibri-1-GGUFCommunity GGUF; fits 64 GB Macs and 2x24 GB. - Q6_K (64.2 GB) / Q8_0 (83.1 GB)
Hob-forge/Kolibri-1-GGUF
Chat template Works
reasoning_effort low/high via chat template; disable with reasoning_effort none or enable_thinking=false. Hermes-style tool calling.
Fix: vLLM: --reasoning-parser kolibri1 --tool-call-parser kolibri1. 1M needs --hf-overrides max_position_embeddings; vendor recommends <=262,144.
Sampling settings
| Official | temp 1.0, top_p 0.97, top_k 128 |
|---|
Runtimes
| vLLM | Works | Official command with kolibri1 parsers. |
|---|---|---|
| llama.cpp | Unverified | Community GGUF exists; architecture support not confirmed. |
| Ollama | Unverified | No library page found. |
| LM Studio | Unverified | |
| MLX | Unverified |
Tool calling in agent tools
| Claude Code (local endpoint) | Unverified | |
|---|---|---|
| OpenCode | Unverified | |
| Cline | Unverified | |
| Aider | Unverified |
Known issues
Sources checked
Spotted something wrong or a new fix? Corrections are reviewed before publishing. (Submission form coming soon.)