Devstral Small 2505 needs about 14.9 GB of memory at Q4_K_M (a 13.1 GB file) with an 8K context window, so it fits on a 16 GB GPU. This repository has 13 quants, from 9.4 GB (Q3_K_S) to 47.2 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 5.4 GB.
Devstral Small 2505 is a 23.6B-parameter Mistral model in GGUF format, quantized by eaddario. Its trained context length is 128K tokens. The repository has 13 quant files, from 9.4 GB (Q3_K_S) to 47.2 GB (F16).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q3_K_S | 9.4 GB | 11.3 GB | Too big | Get file → | |
| IQ3_S | 10.0 GB | 11.9 GB | Too big | Get file → | |
| Q3_K_M | 10.4 GB | 12.3 GB | Too big | Get file → | |
| IQ3_M | 10.8 GB | 12.7 GB | Too big | Get file → | |
| Q3_K_L | 11.4 GB | 13.2 GB | Too big | Get file → | |
| Q4_K_S | 12.2 GB | 14.0 GB | Too big | Get file → | |
| IQ4_NL | 13.1 GB | 14.9 GB | Too big | Get file → | |
| Q4_K_M | 13.1 GB | 14.9 GB | Too big | Get file → | |
| Q5_K_S | 15.1 GB | 17.0 GB | Too big | Get file → | |
| Q5_K_M | 16.0 GB | 17.9 GB | Too big | Get file → | |
| Q6_K | 19.8 GB | 21.6 GB | Too big | Get file → | |
| Q8_0 | 21.2 GB | 23.1 GB | Too big | Get file → | |
| F16 | 47.2 GB | 49.0 GB | Too big | Get file → |
What GPU do you need for Devstral Small 2505?
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 IQ3_S (10.0 GB file, about 11.9 GB in use).
- 16 GB VRAM: the best quant that fits is Q4_K_M (13.1 GB file, about 14.9 GB in use).
- 24 GB VRAM: the best quant that fits is Q8_0 (21.2 GB file, about 23.1 GB in use).
- 32 GB VRAM: the best quant that fits is Q8_0 (21.2 GB file, about 23.1 GB in use).
- 48 GB VRAM: the best quant that fits is Q8_0 (21.2 GB file, about 23.1 GB in use).
Which Devstral Small 2505 quant should you choose?
Q4_K_M is the usual starting point for Devstral Small 2505: it balances file size and quality well. Here it is a 13.1 GB file.
The Q8_0 file is 21.2 GB, about 1.6 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 Q3_K_S at 9.4 GB, with the quality trade-offs that come with fewer bits. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Devstral Small 2505
How much VRAM does Devstral Small 2505 need?
About 14.9 GB at Q4_K_M with an 8K context window. That is the 13.1 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 Devstral Small 2505 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 11.3 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Devstral Small 2505 GGUF?
It depends on the quant: Q3_K_S is 9.4 GB, IQ3_S is 10.0 GB, Q3_K_M is 10.4 GB, IQ3_M is 10.8 GB, Q3_K_L is 11.4 GB, Q4_K_S is 12.2 GB and 7 more.
What context length does Devstral Small 2505 support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 1.3 GB at 8K and 5.4 GB at 32K.
Where can I download Devstral Small 2505 GGUF?
The files are on Hugging Face in the eaddario/Devstral-Small-2505-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 Devstral Small 2505 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.