Kolibri 1 needs about 51.1 GB of memory at Q4_K_M (a 49.8 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 28 quants, from 0.9 GB to 156.3 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 3.4 GB.
Kolibri 1 – a 78B mixture‑of‑experts model for long‑context reasoning in German and English
Kolibri 1 is a 78.1B parameter mixture‑of‑experts model optimized for long‑context inference. It supports explicit reasoning and tool calling, and is trained on sequences up to 262,144 tokens. The model is available in multiple GGUF quantizations for llama.cpp.
Good at
- Long‑context reasoning
- Tool‑calling
- Multi‑step problem solving
Limits
- Requires large VRAM for full model
- No vision support
- License is apache‑2.0
Best for
- Document summarization over large texts
- Agentic workflows needing reasoning
- Custom tool integration
Who it’s for. Developers building conversational assistants or agentic systems that need German and English reasoning. The model is suitable for users who can manage the memory requirements and who prefer a mixture‑of‑experts architecture for efficiency.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.9 GB | 2.2 GB | Fits | Get file → | |
| IQ1_S | 17.6 GB | 19.0 GB | Too big | Get file → | |
| IQ1_M | 19.8 GB | 21.1 GB | Too big | Get file → | |
| IQ2_XXS | 22.8 GB | 24.1 GB | Too big | Get file → | |
| IQ2_XS | 24.3 GB | 25.7 GB | Too big | Get file → | |
| IQ2_S | 25.6 GB | 27.0 GB | Too big | Get file → | |
| IQ2_M | 28.1 GB | 29.4 GB | Too big | Get file → | |
| Q2_K | 29.9 GB | 31.3 GB | Too big | Get file → | |
| IQ3_XXS | 33.7 GB | 35.0 GB | Too big | Get file → | |
| IQ3_XS | 35.5 GB | 36.8 GB | Too big | Get file → | |
| Q3_K_S | 35.9 GB | 37.3 GB | Too big | Get file → | |
| Q3_K_M | 37.6 GB | 39.0 GB | Too big | Get file → | |
| Q3_K_L | 39.8 GB | 41.1 GB | Too big | Get file → | |
| IQ3_M | 40.7 GB | 42.0 GB | Too big | Get file → | |
| IQ4_XS | 43.6 GB | 45.0 GB | Too big | Get file → | |
| Q4_0 | 44.6 GB | 45.9 GB | Too big | Get file → | |
| Q4_K_S | 46.4 GB | 47.7 GB | Too big | Get file → | |
| Q4_1 | 49.2 GB | 50.6 GB | Too big | Get file → | |
| IQ4_NL | 49.8 GB | 51.1 GB | Too big | Get file → | |
| Q4_K_M | 49.8 GB | 51.1 GB | Too big | Get file → | |
| Q4_K_L | 53.5 GB | 54.9 GB | Too big | Get file → | |
| Q5_K_S | 55.8 GB (2 parts) | 57.1 GB | Too big | Get file → | |
| Q5_K_M | 60.3 GB (2 parts) | 61.6 GB | Too big | Get file → | |
| Q6_K_S | 65.8 GB (2 parts) | 67.2 GB | Too big | Get file → | |
| Q6_K | 69.0 GB (2 parts) | 70.3 GB | Too big | Get file → | |
| Q6_K_L | 72.5 GB (2 parts) | 73.9 GB | Too big | Get file → | |
| Q8_0 | 83.1 GB (3 parts) | 84.5 GB | Too big | Get file → | |
| BF16 | 156.3 GB (4 parts) | 157.6 GB | Too big | Get file → |
For 8 GB VRAM, use IQ1_S (17.61 GB) or IQ1_M (19.77 GB). For 16 GB VRAM, Q4_K_M (49.76 GB) or Q4_K_L (53.51 GB) provide a balance of size and performance. For 32 GB VRAM, Q5_K_M (60.3 GB) or Q6_K (69 GB) are suitable. Estimates are approximate.
What GPU do you need for Kolibri 1?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is the main file (0.9 GB file, about 2.2 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (0.9 GB file, about 2.2 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (0.9 GB file, about 2.2 GB in use).
- 12 GB VRAM: the best quant that fits is the main file (0.9 GB file, about 2.2 GB in use).
- 16 GB VRAM: the best quant that fits is the main file (0.9 GB file, about 2.2 GB in use).
- 24 GB VRAM: the best quant that fits is IQ1_M (19.8 GB file, about 21.1 GB in use).
- 32 GB VRAM: the best quant that fits is Q2_K (29.9 GB file, about 31.3 GB in use).
- 48 GB VRAM: the best quant that fits is Q4_K_S (46.4 GB file, about 47.7 GB in use).
Which Kolibri 1 quant should you choose?
Q4_K_M is the usual starting point for Kolibri 1: it balances file size and quality well. Here it is a 49.8 GB file.
The Q8_0 file is 83.1 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is the first one in the table at 0.9 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Kolibri 1
How much VRAM does Kolibri 1 need?
About 51.1 GB at Q4_K_M with an 8K context window. That is the 49.8 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Kolibri 1 on an 8 GB GPU?
Yes, with the smaller files: the 0.9 GB file needs about 2.2 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Kolibri 1 GGUF?
It depends on the quant: file is 0.9 GB, IQ1_S is 17.6 GB, IQ1_M is 19.8 GB, IQ2_XXS is 22.8 GB, IQ2_XS is 24.3 GB, IQ2_S is 25.6 GB and 22 more.
What context length does Kolibri 1 support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.8 GB at 8K and 3.4 GB at 32K.
Where can I download Kolibri 1 GGUF?
The files are on Hugging Face in the bartowski/Kolibri-1-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Kolibri 1 use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Kolibri 1 on 8 GB VRAM?
Use the IQ1_S or IQ1_M quantizations, which are 17.61 GB and 19.77 GB respectively. They fit within 8 GB VRAM when loaded with llama.cpp's memory‑efficient settings.
Does Kolibri 1 support vision tasks?
No, Kolibri 1 is a text‑only model and does not support vision inputs.
What is the native context length of Kolibri 1?
The native context length is 262,144 tokens, but the model can process longer contexts up to 1,048,576 tokens with reduced performance.