PH

Phi 3 medium 128k instruct imatrix

PhiGGUF by duyntnetUpdated 2024-05-23
Download on Hugging Face
Parameters14B
Context128K
Layers40
Architecturephi3
Smallest quant3.0 GB
Quick answer

Phi 3 medium 128k instruct imatrix needs about 10.7 GB of memory at Q4_K_M (a 8.6 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 27 quants, from 3.0 GB (IQ1_S) to 14.8 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 6.7 GB.

Phi 3 medium 128k instruct imatrix – 14 billion parameter LLM

Phi 3 medium 128k instruct imatrix is a 14 billion parameter language model with a 131072 token context window. It is available in multiple GGUF quantizations for llama.cpp. The model supports up to 32064 tokens in its tokenizer.

Good at

  • Large context handling
  • Chat‑style instruction following
  • High token vocabulary

Limits

  • No vision support
  • License type is other
  • Requires recent llama.cpp

Best for

  • Chatbots needing long context
  • Fine‑tuning with custom prompts
  • Research on instruction‑following models

Who it’s for. Developers and researchers who need a medium‑size Phi model with a 128k token context. The model is suitable for applications that require extended dialogue or document comprehension.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S3.0 GB 5.1 GB
Fits Get file →
IQ1_M3.2 GB 5.4 GB
Fits Get file →
IQ2_XXS3.7 GB 5.9 GB
Fits Get file →
IQ2_XS4.1 GB 6.3 GB
Fits Get file →
IQ2_S4.3 GB 6.5 GB
Fits Get file →
IQ2_M4.7 GB 6.9 GB
Fits Get file →
Q2_K_S4.8 GB 6.9 GB
Fits Get file →
Q2_K5.1 GB 7.3 GB
Tight Get file →
IQ3_XXS5.5 GB 7.6 GB
Tight Get file →
IQ3_XS5.8 GB 8.0 GB
Tight Get file →
IQ3_S6.1 GB 8.2 GB
Too big Get file →
Q3_K_S6.1 GB 8.2 GB
Too big Get file →
IQ3_M6.5 GB 8.7 GB
Too big Get file →
Q3_K_M6.9 GB 9.1 GB
Too big Get file →
IQ4_XS7.5 GB 9.6 GB
Too big Get file →
Q3_K_L7.5 GB 9.7 GB
Too big Get file →
IQ4_NL7.9 GB 10.1 GB
Too big Get file →
Q4_07.9 GB 10.1 GB
Too big Get file →
Q4_K_S8.0 GB 10.1 GB
Too big Get file →
Q4_K_M8.6 GB 10.7 GB
Too big Get file →
Q4_18.8 GB 10.9 GB
Too big Get file →
Q5_K_S9.6 GB 11.8 GB
Too big Get file →
Q5_09.7 GB 11.8 GB
Too big Get file →
Q5_K_M10.1 GB 12.3 GB
Too big Get file →
Q5_110.5 GB 12.7 GB
Too big Get file →
Q6_K11.5 GB 13.6 GB
Too big Get file →
Q8_014.8 GB 17.0 GB
Too big Get file →

For 8 GB VRAM, use IQ1_S (5.1 GB memory at 8k context). For 12 GB VRAM, IQ3_S (8.2 GB) or Q3_K_S (8.2 GB) are good choices. For 16 GB VRAM, Q4_0 (10.1 GB) or Q5_K_S (11.8 GB) provide higher quality with acceptable memory usage. Estimates are approximate.

What GPU do you need for Phi 3 medium 128k instruct imatrix?

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: the best quant that fits is IQ2_XXS (3.7 GB file, about 5.9 GB in use).
  • 8 GB VRAM: the best quant that fits is IQ3_XS (5.8 GB file, about 8.0 GB in use).
  • 12 GB VRAM: the best quant that fits is Q5_0 (9.7 GB file, about 11.8 GB in use).
  • 16 GB VRAM: the best quant that fits is Q6_K (11.5 GB file, about 13.6 GB in use).
  • 24 GB VRAM: the best quant that fits is Q8_0 (14.8 GB file, about 17.0 GB in use).

Which Phi 3 medium 128k instruct imatrix quant should you choose?

Q4_K_M is the usual starting point for Phi 3 medium 128k instruct imatrix: it balances file size and quality well. Here it is a 8.6 GB file.

The Q8_0 file is 14.8 GB, about 1.7 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 IQ1_S at 3.0 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

Architecturephi3
Parameters14B
Layers40
Trained context131,072 tokens
KV cache (fp16)200 KB per token

Source

Quantized byduyntnet
FamilyPhi
Licenseother
Downloads (HF)1,988
Last updated2024-05-23

Frequently asked questions about Phi 3 medium 128k instruct imatrix

How much VRAM does Phi 3 medium 128k instruct imatrix need?

About 10.7 GB at Q4_K_M with an 8K context window. That is the 8.6 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 Phi 3 medium 128k instruct imatrix on an 8 GB GPU?

Yes, with IQ3_XS: the 5.8 GB file needs about 8.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Phi 3 medium 128k instruct imatrix GGUF?

It depends on the quant: IQ1_S is 3.0 GB, IQ1_M is 3.2 GB, IQ2_XXS is 3.7 GB, IQ2_XS is 4.1 GB, IQ2_S is 4.3 GB, IQ2_M is 4.7 GB and 21 more.

What context length does Phi 3 medium 128k instruct imatrix support?

The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 1.7 GB at 8K and 6.7 GB at 32K.

Where can I download Phi 3 medium 128k instruct imatrix GGUF?

The files are on Hugging Face in the duyntnet/Phi-3-medium-128k-instruct-imatrix-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 medium 128k instruct imatrix use?

Hugging Face lists the license as other. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Can I run Phi 3 medium 128k instruct imatrix on 8 GB VRAM?

Yes, the IQ1_S quantization requires about 5.1 GB of VRAM at 8k context, which fits within an 8 GB GPU.

Does this model support image inputs?

No, the model does not have vision capabilities.

What is the maximum context length I can use?

The model was trained with 131072 tokens and can handle that context length in inference.

Similar models

Browse all →

More from duyntnet

All by duyntnet →

More Phi models

All Phi →

Other medium models

All medium models →