Phi 3.5 MoE Instruct needs about 26.9 GB of memory at Q4_K_M (a 25.3 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 11 quants, from 15.3 GB (Q2_K) to 44.5 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.
Phi 3.5 MoE Instruct is a 41.9B-parameter Phi model in GGUF format, quantized by mradermacher. Its trained context length is 128K tokens. The repository has 11 quant files, from 15.3 GB (Q2_K) to 44.5 GB (Q8_0).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 15.3 GB | 16.8 GB | Too big | Get file → | |
| Q3_K_S | 18.1 GB | 19.6 GB | Too big | Get file → | |
| Q3_K_M | 20.0 GB | 21.6 GB | Too big | Get file → | |
| Q3_K_L | 21.7 GB | 23.3 GB | Too big | Get file → | |
| IQ4_XS | 22.6 GB | 24.2 GB | Too big | Get file → | |
| Q4_K_S | 23.8 GB | 25.4 GB | Too big | Get file → | |
| Q4_K_M | 25.3 GB | 26.9 GB | Too big | Get file → | |
| Q5_K_S | 28.8 GB | 30.4 GB | Too big | Get file → | |
| Q5_K_M | 29.7 GB | 31.3 GB | Too big | Get file → | |
| Q6_K | 34.4 GB | 35.9 GB | Too big | Get file → | |
| Q8_0 | 44.5 GB | 46.1 GB | Too big | Get file → |
What GPU do you need for Phi 3.5 MoE Instruct?
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: the best quant that fits is Q3_K_L (21.7 GB file, about 23.3 GB in use).
- 32 GB VRAM: the best quant that fits is Q5_K_M (29.7 GB file, about 31.3 GB in use).
- 48 GB VRAM: the best quant that fits is Q8_0 (44.5 GB file, about 46.1 GB in use).
Which Phi 3.5 MoE Instruct quant should you choose?
Q4_K_M is the usual starting point for Phi 3.5 MoE Instruct: it balances file size and quality well. Here it is a 25.3 GB file.
The Q8_0 file is 44.5 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 Q2_K at 15.3 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 Phi 3.5 MoE Instruct
How much VRAM does Phi 3.5 MoE Instruct need?
About 26.9 GB at Q4_K_M with an 8K context window. That is the 25.3 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 Phi 3.5 MoE Instruct on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 16.8 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Phi 3.5 MoE Instruct GGUF?
It depends on the quant: Q2_K is 15.3 GB, Q3_K_S is 18.1 GB, Q3_K_M is 20.0 GB, Q3_K_L is 21.7 GB, IQ4_XS is 22.6 GB, Q4_K_S is 23.8 GB and 5 more.
What context length does Phi 3.5 MoE Instruct 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 Phi 3.5 MoE Instruct GGUF?
The files are on Hugging Face in the mradermacher/Phi-3.5-MoE-instruct-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 Phi 3.5 MoE Instruct use?
Hugging Face lists the license as mit. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.