PH

Phi 2 OmniMatrix

PhiGGUF by mradermacherUpdated 2024-09-15
Download on Hugging Face
Parameters2.78B
Context2K
Layers32
Architecturephi2
Smallest quant1.1 GB
Quick answer

Phi 2 OmniMatrix needs about 4.9 GB of memory at Q4_K_M (a 1.7 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 15 quants, from 1.1 GB (Q2_K) to 5.6 GB (F16).

Phi 2 OmniMatrix – 2.78B parameter language model

Phi 2 OmniMatrix is a 2.78B parameter model built on the phi2 architecture. It merges several pre‑trained models using task arithmetic merge. The model supports a 2048 token context window.

Good at

  • Small footprint variants
  • Efficient inference on modest GPUs
  • Versatile merge of multiple task models

Limits

  • No vision support
  • License not specified
  • Context limited to 2048 tokens

Best for

  • Chat applications on 8‑GB VRAM
  • Lightweight code instruction tasks
  • Multi‑model merged inference

Who it’s for. Developers and researchers needing a compact Phi‑2 variant that can run on consumer GPUs. The model is suitable for those who want to experiment with merged task models without large memory requirements.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K1.1 GB 2.3 GB
Fits Get file →
IQ3_XS1.2 GB 2.4 GB
Fits Get file →
IQ3_S1.3 GB 2.4 GB
Fits Get file →
Q3_K_S1.3 GB 2.4 GB
Fits Get file →
IQ3_M1.3 GB 2.5 GB
Fits Get file →
Q3_K_M1.4 GB 2.6 GB
Fits Get file →
IQ4_XS1.5 GB 2.7 GB
Fits Get file →
Q3_K_L1.6 GB 2.7 GB
Fits Get file →
Q4_K_S1.6 GB 2.8 GB
Fits Get file →
Q4_K_M1.7 GB 2.9 GB
Fits Get file →
Q5_K_S1.9 GB 3.1 GB
Fits Get file →
Q5_K_M2.0 GB 3.2 GB
Fits Get file →
Q6_K2.3 GB 3.5 GB
Fits Get file →
Q8_03.0 GB 4.1 GB
Fits Get file →
F165.6 GB 6.7 GB
Fits Get file →

For 8‑GB VRAM use Q4_K_M (1.74 GB file, ~4.9 GB memory). If 6‑GB VRAM is available, Q3_K_M (1.43 GB file, ~4.6 GB memory) is a good choice. For 4‑GB VRAM consider Q2_K (1.11 GB file, ~4.3 GB memory).

What GPU do you need for Phi 2 OmniMatrix?

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 Q6_K (2.3 GB file, about 5.5 GB in use).
  • 8 GB VRAM: the best quant that fits is Q8_0 (3.0 GB file, about 6.1 GB in use).
  • 12 GB VRAM: the best quant that fits is F16 (5.6 GB file, about 8.7 GB in use).

Which Phi 2 OmniMatrix quant should you choose?

Q4_K_M is the usual starting point for Phi 2 OmniMatrix: it balances file size and quality well. Here it is a 1.7 GB file.

The Q8_0 file is 3.0 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 Q2_K at 1.1 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

Architecturephi2
Parameters2.78B
Layers32
Trained context2,048 tokens
KV cache (fp16)320 KB per token

Source

Quantized bymradermacher
FamilyPhi
Downloads (HF)1,790
Last updated2024-09-15

Frequently asked questions about Phi 2 OmniMatrix

How much VRAM does Phi 2 OmniMatrix need?

About 4.9 GB at Q4_K_M with an 8K context window. That is the 1.7 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 2 OmniMatrix on an 8 GB GPU?

Yes, with Q8_0: the 3.0 GB file needs about 6.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Phi 2 OmniMatrix GGUF?

It depends on the quant: Q2_K is 1.1 GB, IQ3_XS is 1.2 GB, IQ3_S is 1.3 GB, Q3_K_S is 1.3 GB, IQ3_M is 1.3 GB, Q3_K_M is 1.4 GB and 9 more.

What context length does Phi 2 OmniMatrix support?

The GGUF header lists a trained context length of 2,048 tokens (2K). The KV cache grows with the context you use: about 2.7 GB at 8K and 0.7 GB at 2K.

Where can I download Phi 2 OmniMatrix GGUF?

The files are on Hugging Face in the mradermacher/Phi-2-OmniMatrix-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

Can I run Phi 2 OmniMatrix on 8 GB VRAM?

Yes, the Q4_K_M quant uses about 4.9 GB of VRAM at 8k context, fitting within an 8 GB GPU.

Does this model support vision tasks?

No, the model has no vision capabilities.

What is the maximum context length?

The model was trained with a 2048 token context window.

Other GGUFs of Phi-2-OmniMatrix

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Phi models

All Phi →

Other tiny models

All tiny models →