LLaMA 1 65B needs about 61.3 GB of memory at Q4_K_M (a 39.3 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 16 quants, from 14.7 GB (TQ1_0) to 130.6 GB (F16).
LLaMA 1 65B is a 65.3B-parameter Llama model in GGUF format, quantized by PhilLovesCats. Its trained context length is 2K tokens. The repository has 16 quant files, from 14.7 GB (TQ1_0) to 130.6 GB (F16).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| TQ1_0 | 14.7 GB | 20.6 GB | Too big | Get file → | |
| TQ2_0 | 17.8 GB | 23.6 GB | Too big | Get file → | |
| Q2_K | 24.1 GB | 30.0 GB | Too big | Get file → | |
| Q3_K_S | 28.2 GB | 34.0 GB | Too big | Get file → | |
| Q3_K_M | 31.6 GB | 37.5 GB | Too big | Get file → | |
| Q3_K_L | 34.6 GB | 40.5 GB | Too big | Get file → | |
| Q4_0 | 36.8 GB | 42.7 GB | Too big | Get file → | |
| Q4_K_S | 37.1 GB | 42.9 GB | Too big | Get file → | |
| Q4_K_M | 39.3 GB | 45.2 GB | Too big | Get file → | |
| Q5_0 | 44.9 GB | 50.8 GB | Too big | Get file → | |
| Q5_K_S | 44.9 GB | 50.8 GB | Too big | Get file → | |
| Q5_K_M | 46.2 GB | 52.1 GB | Too big | Get file → | |
| Q6_K | 53.6 GB (6 parts) | 59.4 GB | Too big | Get file → | |
| Q8_0 | 69.4 GB (6 parts) | 75.2 GB | Too big | Get file → | |
| BF16 | 130.6 GB (6 parts) | 136.4 GB | Too big | Get file → | |
| F16 | 130.6 GB (6 parts) | 136.4 GB | Too big | Get file → |
What GPU do you need for LLaMA 1 65B?
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: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: no quant fits at an 8K context.
- 32 GB VRAM: no quant fits at an 8K context.
- 48 GB VRAM: the best quant that fits is Q2_K (24.1 GB file, about 46.1 GB in use).
Which LLaMA 1 65B quant should you choose?
Q4_K_M is the usual starting point for LLaMA 1 65B: it balances file size and quality well. Here it is a 39.3 GB file.
The Q8_0 file is 69.4 GB, about 1.8 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 TQ1_0 at 14.7 GB, with the quality trade-offs that come with fewer bits. Lower bit counts mean smaller files and more quality loss.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about LLaMA 1 65B
How much VRAM does LLaMA 1 65B need?
About 61.3 GB at Q4_K_M with an 8K context window. That is the 39.3 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 LLaMA 1 65B on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 36.7 GB, so you would need a larger GPU or CPU offloading.
What is the file size of LLaMA 1 65B GGUF?
It depends on the quant: TQ1_0 is 14.7 GB, TQ2_0 is 17.8 GB, Q2_K is 24.1 GB, Q3_K_S is 28.2 GB, Q3_K_M is 31.6 GB, Q3_K_L is 34.6 GB and 10 more.
What context length does LLaMA 1 65B support?
The GGUF header lists a trained context length of 2,048 tokens (2K). The KV cache grows with the context you use: about 21.5 GB at 8K and 5.4 GB at 2K.
Where can I download LLaMA 1 65B GGUF?
The files are on Hugging Face in the PhilLovesCats/LLaMA-1-65B-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.