Umbra v2.1 MoE 4x10.7 needs about 23.9 GB of memory at Q4_K_M (a 21.8 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 10 quants, from 9.5 GB (IQ2_XXS) to 29.6 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 6.4 GB.
Umbra v2.1 MoE 4x10.7 is a 36.1B-parameter Umbra model in GGUF format, quantized by LoneStriker. Its trained context length is 32K tokens. The repository has 10 quant files, from 9.5 GB (IQ2_XXS) to 29.6 GB (Q6_K).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ2_XXS | 9.5 GB | 11.6 GB | Too big | Get file → | |
| Q2_K | 13.2 GB | 15.3 GB | Too big | Get file → | |
| Q3_K_S | 15.6 GB | 17.7 GB | Too big | Get file → | |
| Q3_K_M | 17.3 GB | 19.4 GB | Too big | Get file → | |
| Q3_K_L | 18.7 GB | 20.8 GB | Too big | Get file → | |
| Q4_K_S | 20.5 GB | 22.6 GB | Too big | Get file → | |
| Q4_K_M | 21.8 GB | 23.9 GB | Too big | Get file → | |
| Q5_K_S | 24.8 GB | 26.9 GB | Too big | Get file → | |
| Q5_K_M | 25.6 GB | 27.7 GB | Too big | Get file → | |
| Q6_K | 29.6 GB | 31.7 GB | Too big | Get file → |
What GPU do you need for Umbra v2.1 MoE 4x10.7?
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: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: the best quant that fits is IQ2_XXS (9.5 GB file, about 11.6 GB in use).
- 16 GB VRAM: the best quant that fits is Q2_K (13.2 GB file, about 15.3 GB in use).
- 24 GB VRAM: the best quant that fits is Q4_K_M (21.8 GB file, about 23.9 GB in use).
- 32 GB VRAM: the best quant that fits is Q6_K (29.6 GB file, about 31.7 GB in use).
Which Umbra v2.1 MoE 4x10.7 quant should you choose?
Q4_K_M is the usual starting point for Umbra v2.1 MoE 4x10.7: it balances file size and quality well. Here it is a 21.8 GB file.
If memory is tight, the smallest file is IQ2_XXS at 9.5 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Umbra v2.1 MoE 4x10.7
How much VRAM does Umbra v2.1 MoE 4x10.7 need?
About 23.9 GB at Q4_K_M with an 8K context window. That is the 21.8 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 Umbra v2.1 MoE 4x10.7 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 11.6 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Umbra v2.1 MoE 4x10.7 GGUF?
It depends on the quant: IQ2_XXS is 9.5 GB, Q2_K is 13.2 GB, Q3_K_S is 15.6 GB, Q3_K_M is 17.3 GB, Q3_K_L is 18.7 GB, Q4_K_S is 20.5 GB and 4 more.
What context length does Umbra v2.1 MoE 4x10.7 support?
The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 1.6 GB at 8K and 6.4 GB at 32K.
Where can I download Umbra v2.1 MoE 4x10.7 GGUF?
The files are on Hugging Face in the LoneStriker/Umbra-v2.1-MoE-4x10.7-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 Umbra v2.1 MoE 4x10.7 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.