Llama3.1 16B Upscaled needs about 10.9 GB of memory at IQ4_XS (a 8.3 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 9 quants, from 5.7 GB (Q2_K) to 16.0 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 8.6 GB.
Llama3.1 16B Upscaled is a 15B-parameter Llama model in GGUF format, quantized by mradermacher. Its trained context length is 128K tokens. The repository has 9 quant files, from 5.7 GB (Q2_K) to 16.0 GB (Q8_0).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 5.7 GB | 8.4 GB | Too big | Get file → | |
| Q3_K_S | 6.7 GB | 9.3 GB | Too big | Get file → | |
| Q3_K_M | 7.4 GB | 10.0 GB | Too big | Get file → | |
| Q3_K_L | 8.0 GB | 10.6 GB | Too big | Get file → | |
| IQ4_XS | 8.3 GB | 10.9 GB | Too big | Get file → | |
| Q4_K_S | 8.6 GB | 11.3 GB | Too big | Get file → | |
| Q5_K_S | 10.4 GB | 13.0 GB | Too big | Get file → | |
| Q6_K | 12.3 GB | 15.0 GB | Too big | Get file → | |
| Q8_0 | 16.0 GB | 18.6 GB | Too big | Get file → |
What GPU do you need for Llama3.1 16B Upscaled?
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 Q4_K_S (8.6 GB file, about 11.3 GB in use).
- 16 GB VRAM: the best quant that fits is Q6_K (12.3 GB file, about 15.0 GB in use).
- 24 GB VRAM: the best quant that fits is Q8_0 (16.0 GB file, about 18.6 GB in use).
Which Llama3.1 16B Upscaled quant should you choose?
IQ4_XS (8.3 GB) is the middle option in this repository. The usual sweet spot between file size and quality, and the quant most people start with.
If memory is tight, the smallest file is Q2_K at 5.7 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 Llama3.1 16B Upscaled
How much VRAM does Llama3.1 16B Upscaled need?
About 10.9 GB at IQ4_XS with an 8K context window. That is the 8.3 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 Llama3.1 16B Upscaled on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 8.4 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Llama3.1 16B Upscaled GGUF?
It depends on the quant: Q2_K is 5.7 GB, Q3_K_S is 6.7 GB, Q3_K_M is 7.4 GB, Q3_K_L is 8.0 GB, IQ4_XS is 8.3 GB, Q4_K_S is 8.6 GB and 3 more.
What context length does Llama3.1 16B Upscaled support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 2.1 GB at 8K and 8.6 GB at 32K.
Where can I download Llama3.1 16B Upscaled GGUF?
The files are on Hugging Face in the mradermacher/Llama3.1-16B-Upscaled-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.