ontocord phi 3 22b 128k needs about 16.8 GB of memory at Q4_K_M (a 13.6 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 18 quants, from 8.1 GB (Q2_K) to 16.0 GB (Q5_K). Longer context needs more: at 32K tokens the KV cache alone adds about 10.7 GB.
ontocord phi 3 22b 128k is a 22.1B-parameter Phi model in GGUF format, quantized by RichardErkhov. Its trained context length is 128K tokens. The repository has 18 quant files, from 8.1 GB (Q2_K) to 16.0 GB (Q5_K).
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q2_K | 8.1 GB | 11.3 GB | Too big | Get file → | |
| IQ3_XS | 9.2 GB | 12.4 GB | Too big | Get file → | |
| IQ3_S | 9.6 GB | 12.8 GB | Too big | Get file → | |
| Q3_K_S | 9.6 GB | 12.8 GB | Too big | Get file → | |
| IQ3_M | 10.2 GB | 13.4 GB | Too big | Get file → | |
| Q3_K | 11.0 GB | 14.1 GB | Too big | Get file → | |
| Q3_K_M | 11.0 GB | 14.1 GB | Too big | Get file → | |
| Q3_K_L | 11.9 GB | 15.0 GB | Too big | Get file → | |
| IQ4_XS | 11.9 GB | 15.1 GB | Too big | Get file → | |
| Q4_0 | 12.5 GB | 15.7 GB | Too big | Get file → | |
| IQ4_NL | 12.6 GB | 15.8 GB | Too big | Get file → | |
| Q4_K_S | 12.6 GB | 15.8 GB | Too big | Get file → | |
| Q4_K | 13.6 GB | 16.8 GB | Too big | Get file → | |
| Q4_K_M | 13.6 GB | 16.8 GB | Too big | Get file → | |
| Q4_1 | 13.9 GB | 17.1 GB | Too big | Get file → | |
| Q5_0 | 15.2 GB | 18.4 GB | Too big | Get file → | |
| Q5_K_S | 15.2 GB | 18.4 GB | Too big | Get file → | |
| Q5_K | 16.0 GB | 19.2 GB | Too big | Get file → |
What GPU do you need for ontocord phi 3 22b 128k?
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 Q2_K (8.1 GB file, about 11.3 GB in use).
- 16 GB VRAM: the best quant that fits is IQ4_NL (12.6 GB file, about 15.8 GB in use).
- 24 GB VRAM: the best quant that fits is Q5_K (16.0 GB file, about 19.2 GB in use).
Which ontocord phi 3 22b 128k quant should you choose?
Q4_K_M is the usual starting point for ontocord phi 3 22b 128k: it balances file size and quality well. Here it is a 13.6 GB file.
If memory is tight, the smallest file is Q2_K at 8.1 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
Frequently asked questions about ontocord phi 3 22b 128k
How much VRAM does ontocord phi 3 22b 128k need?
About 16.8 GB at Q4_K_M with an 8K context window. That is the 13.6 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 ontocord phi 3 22b 128k on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 11.3 GB, so you would need a larger GPU or CPU offloading.
What is the file size of ontocord phi 3 22b 128k GGUF?
It depends on the quant: Q2_K is 8.1 GB, IQ3_XS is 9.2 GB, IQ3_S is 9.6 GB, Q3_K_S is 9.6 GB, IQ3_M is 10.2 GB, Q3_K is 11.0 GB and 12 more.
What context length does ontocord phi 3 22b 128k support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 2.7 GB at 8K and 10.7 GB at 32K.
Where can I download ontocord phi 3 22b 128k GGUF?
The files are on Hugging Face in the RichardErkhov/ontocord_-_phi-3-22b-128k-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 ontocord phi 3 22b 128k 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.