Falcon 40b needs about 27.0 GB of memory at Q4_K_M (a 25.5 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 23 quants, from 9.2 GB (IQ1_S) to 34.5 GB (Q6_K).
Falcon 40b is a 41.8B-parameter Falcon model in GGUF format, quantized by mradermacher. Its trained context length is 2K tokens. The repository has 23 quant files, from 9.2 GB (IQ1_S) to 34.5 GB (Q6_K).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ1_S | 9.2 GB | 10.0 GB | Too big | Get file → | |
| IQ1_M | 10.0 GB | 10.8 GB | Too big | Get file → | |
| IQ2_XXS | 11.4 GB | 12.1 GB | Too big | Get file → | |
| IQ2_XS | 12.6 GB | 13.4 GB | Too big | Get file → | |
| IQ2_S | 13.4 GB | 14.2 GB | Too big | Get file → | |
| IQ2_M | 14.5 GB | 15.3 GB | Too big | Get file → | |
| Q2_K_S | 14.6 GB | 15.3 GB | Too big | Get file → | |
| Q2_K | 15.8 GB | 16.5 GB | Too big | Get file → | |
| IQ3_XXS | 16.4 GB | 17.2 GB | Too big | Get file → | |
| IQ3_XS | 17.8 GB | 18.5 GB | Too big | Get file → | |
| IQ3_S | 18.3 GB | 19.1 GB | Too big | Get file → | |
| Q3_K_S | 18.3 GB | 19.1 GB | Too big | Get file → | |
| IQ3_M | 19.2 GB | 19.9 GB | Too big | Get file → | |
| Q3_K_M | 20.1 GB | 20.8 GB | Too big | Get file → | |
| Q3_K_L | 21.6 GB | 22.4 GB | Too big | Get file → | |
| IQ4_XS | 22.5 GB | 23.3 GB | Too big | Get file → | |
| Q4_K_S | 23.8 GB | 24.6 GB | Too big | Get file → | |
| Q4_0 | 23.9 GB | 24.7 GB | Too big | Get file → | |
| Q4_K_M | 25.5 GB | 26.2 GB | Too big | Get file → | |
| Q4_1 | 26.4 GB | 27.1 GB | Too big | Get file → | |
| Q5_K_S | 29.0 GB | 29.7 GB | Too big | Get file → | |
| Q5_K_M | 30.6 GB | 31.4 GB | Too big | Get file → | |
| Q6_K | 34.5 GB | 35.2 GB | Too big | Get file → |
What GPU do you need for Falcon 40b?
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 IQ1_M (10.0 GB file, about 11.5 GB in use).
- 16 GB VRAM: the best quant that fits is IQ2_M (14.5 GB file, about 16.0 GB in use).
- 24 GB VRAM: the best quant that fits is IQ4_XS (22.5 GB file, about 24.0 GB in use).
- 32 GB VRAM: the best quant that fits is Q5_K_S (29.0 GB file, about 30.5 GB in use).
- 48 GB VRAM: the best quant that fits is Q6_K (34.5 GB file, about 36.0 GB in use).
Which Falcon 40b quant should you choose?
Q4_K_M is the usual starting point for Falcon 40b: it balances file size and quality well. Here it is a 25.5 GB file.
If memory is tight, the smallest file is IQ1_S at 9.2 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 Falcon 40b
How much VRAM does Falcon 40b need?
About 27.0 GB at Q4_K_M with an 8K context window. That is the 25.5 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 Falcon 40b on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 10.7 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Falcon 40b GGUF?
It depends on the quant: IQ1_S is 9.2 GB, IQ1_M is 10.0 GB, IQ2_XXS is 11.4 GB, IQ2_XS is 12.6 GB, IQ2_S is 13.4 GB, IQ2_M is 14.5 GB and 17 more.
What context length does Falcon 40b support?
The GGUF header lists a trained context length of 2,048 tokens (2K). The KV cache grows with the context you use: about 1.0 GB at 8K and 0.3 GB at 2K.
Where can I download Falcon 40b GGUF?
The files are on Hugging Face in the mradermacher/falcon-40b-i1-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 Falcon 40b 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.