Llama 3.1 8B needs about 6.5 GB of memory at Q4_K_M (a 4.9 GB file) with an 8K context window, so it fits on a 8 GB GPU. This repository has 22 quants, from 0.4 GB (IQ3_S) to 8.5 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.
Llama 3.1 8B is a 8.03B-parameter Llama model in GGUF format, quantized by RichardErkhov. Its trained context length is 128K tokens. The repository has 22 quant files, from 0.4 GB (IQ3_S) to 8.5 GB (Q8_0).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ3_S | 0.4 GB | 2.0 GB | Fits | Get file → | |
| Q2_K | 3.2 GB | 4.8 GB | Fits | Get file → | |
| IQ3_XS | 3.5 GB | 5.1 GB | Fits | Get file → | |
| Q3_K_S | 3.7 GB | 5.2 GB | Fits | Get file → | |
| IQ3_M | 3.8 GB | 5.4 GB | Fits | Get file → | |
| Q3_K | 4.0 GB | 5.6 GB | Fits | Get file → | |
| Q3_K_M | 4.0 GB | 5.6 GB | Fits | Get file → | |
| Q3_K_L | 4.3 GB | 5.9 GB | Fits | Get file → | |
| IQ4_XS | 4.5 GB | 6.1 GB | Fits | Get file → | |
| Q4_0 | 4.7 GB | 6.2 GB | Fits | Get file → | |
| Q4_K_S | 4.7 GB | 6.3 GB | Fits | Get file → | |
| IQ4_NL | 4.7 GB | 6.3 GB | Fits | Get file → | |
| Q4_K | 4.9 GB | 6.5 GB | Fits | Get file → | |
| Q4_K_M | 4.9 GB | 6.5 GB | Fits | Get file → | |
| Q4_1 | 5.1 GB | 6.7 GB | Fits | Get file → | |
| Q5_0 | 5.6 GB | 7.2 GB | Fits | Get file → | |
| Q5_K_S | 5.6 GB | 7.2 GB | Fits | Get file → | |
| Q5_K | 5.7 GB | 7.3 GB | Tight | Get file → | |
| Q5_K_M | 5.7 GB | 7.3 GB | Tight | Get file → | |
| Q5_1 | 6.1 GB | 7.6 GB | Tight | Get file → | |
| Q6_K | 6.6 GB | 8.2 GB | Too big | Get file → | |
| Q8_0 | 8.5 GB | 10.1 GB | Too big | Get file → |
What GPU do you need for Llama 3.1 8B?
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 IQ3_S (0.4 GB file, about 2.0 GB in use).
- 6 GB VRAM: the best quant that fits is Q3_K_L (4.3 GB file, about 5.9 GB in use).
- 8 GB VRAM: the best quant that fits is Q5_1 (6.1 GB file, about 7.6 GB in use).
- 12 GB VRAM: the best quant that fits is Q8_0 (8.5 GB file, about 10.1 GB in use).
Which Llama 3.1 8B quant should you choose?
Q4_K_M is the usual starting point for Llama 3.1 8B: it balances file size and quality well. Here it is a 4.9 GB file.
The Q8_0 file is 8.5 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 IQ3_S at 0.4 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
Frequently asked questions about Llama 3.1 8B
How much VRAM does Llama 3.1 8B need?
About 6.5 GB at Q4_K_M with an 8K context window. That is the 4.9 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 Llama 3.1 8B on an 8 GB GPU?
Yes, with Q5_1: the 6.1 GB file needs about 7.6 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Llama 3.1 8B GGUF?
It depends on the quant: IQ3_S is 0.4 GB, Q2_K is 3.2 GB, IQ3_XS is 3.5 GB, Q3_K_S is 3.7 GB, IQ3_M is 3.8 GB, Q3_K is 4.0 GB and 16 more.
What context length does Llama 3.1 8B support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.
Where can I download Llama 3.1 8B GGUF?
The files are on Hugging Face in the RichardErkhov/daliabess7_-_Llama-3.1-8B-python_update-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 Llama 3.1 8B 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.