PH

phi3-4x4b-v1

PhiGGUF by RichardErkhovUpdated 2024-06-04
Download on Hugging Face
Parameters11.1B
Context4K
Layers32
Architecturellama
Smallest quant4.1 GB
Quick answer

phi3-4x4b-v1 needs about 10.4 GB of memory at Q4_K_M (a 6.7 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 22 quants, from 4.1 GB (Q2_K) to 11.8 GB (Q8_0).

Phi3‑4x4b‑v1: 11.1B sparse‑MOE LLaMA‑style model

Phi3‑4x4b‑v1 is an 11.1B parameter sparse‑MOE model based on the Phi family. It was trained on 4096‑token context and is not fine‑tuned on instruct data. The model is available in multiple GGUF quantizations.

Good at

  • Sparse‑MOE inference
  • Large context handling
  • Low‑precision deployment

Limits

  • No vision support
  • No license specified
  • Not trained on instruct data

Best for

  • Local LLM inference
  • Research on sparse‑MOE models
  • Low‑memory deployment

Who it’s for. Developers and researchers who need a lightweight Phi‑style model for local inference. The model is suitable for experimentation with sparse‑MOE architectures and low‑precision quantization.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K4.1 GB 6.2 GB
Fits Get file →
IQ3_XS4.5 GB 6.7 GB
Fits Get file →
IQ3_S4.8 GB 6.9 GB
Fits Get file →
Q3_K_S4.8 GB 6.9 GB
Fits Get file →
IQ3_M4.9 GB 7.0 GB
Fits Get file →
Q3_K5.3 GB 7.4 GB
Tight Get file →
Q3_K_M5.3 GB 7.4 GB
Tight Get file →
Q3_K_L5.8 GB 7.9 GB
Tight Get file →
IQ4_XS6.0 GB 8.1 GB
Too big Get file →
Q4_06.3 GB 8.4 GB
Too big Get file →
IQ4_NL6.3 GB 8.4 GB
Too big Get file →
Q4_K_S6.3 GB 8.4 GB
Too big Get file →
Q4_K6.7 GB 8.8 GB
Too big Get file →
Q4_K_M6.7 GB 8.8 GB
Too big Get file →
Q4_16.9 GB 9.1 GB
Too big Get file →
Q5_07.6 GB 9.7 GB
Too big Get file →
Q5_K_S7.6 GB 9.7 GB
Too big Get file →
Q5_K7.9 GB 10.0 GB
Too big Get file →
Q5_K_M7.9 GB 10.0 GB
Too big Get file →
Q5_18.3 GB 10.4 GB
Too big Get file →
Q6_K9.1 GB 11.2 GB
Too big Get file →
Q8_011.8 GB 13.9 GB
Too big Get file →

For 8 GB VRAM, use Q4_0 (6.25 GB file, ~10 GB memory). For 12 GB VRAM, Q5_0 (7.63 GB file, ~11.3 GB memory) or Q5_K (7.86 GB file, ~11.6 GB memory). For 16 GB VRAM, Q6_K (9.08 GB file, ~12.8 GB memory) or Q8_0 (11.76 GB file, ~15.5 GB memory).

What GPU do you need for phi3-4x4b-v1?

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: the best quant that fits is Q2_K (4.1 GB file, about 7.8 GB in use).
  • 12 GB VRAM: the best quant that fits is Q5_1 (8.3 GB file, about 12.0 GB in use).
  • 16 GB VRAM: the best quant that fits is Q8_0 (11.8 GB file, about 15.5 GB in use).

Which phi3-4x4b-v1 quant should you choose?

Q4_K_M is the usual starting point for phi3-4x4b-v1: it balances file size and quality well. Here it is a 6.7 GB file.

The Q8_0 file is 11.8 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 4.1 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

Architecturellama
Parameters11.1B (approximate, mixture of experts)
Layers32
Trained context4,096 tokens
KV cache (fp16)384 KB per token

Source

Quantized byRichardErkhov
FamilyPhi
Downloads (HF)1,127
Last updated2024-06-04

Frequently asked questions about phi3-4x4b-v1

How much VRAM does phi3-4x4b-v1 need?

About 10.4 GB at Q4_K_M with an 8K context window. That is the 6.7 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 phi3-4x4b-v1 on an 8 GB GPU?

Yes, with Q2_K: the 4.1 GB file needs about 7.8 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of phi3-4x4b-v1 GGUF?

It depends on the quant: Q2_K is 4.1 GB, IQ3_XS is 4.5 GB, IQ3_S is 4.8 GB, Q3_K_S is 4.8 GB, IQ3_M is 4.9 GB, Q3_K is 5.3 GB and 16 more.

What context length does phi3-4x4b-v1 support?

The GGUF header lists a trained context length of 4,096 tokens (4K). The KV cache grows with the context you use: about 3.2 GB at 8K and 1.6 GB at 4K.

Where can I download phi3-4x4b-v1 GGUF?

The files are on Hugging Face in the RichardErkhov/Fizzarolli_-_phi3-4x4b-v1-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 phi3-4x4b-v1 on 8 GB VRAM?

Yes, using the Q4_0 quantization which requires about 10 GB of VRAM at 8k context.

Is this model licensed for commercial use?

The license is not specified in the provided data.

Does phi3-4x4b-v1 support vision tasks?

No, the model has no vision capabilities.

Similar models

Browse all →

More from RichardErkhov

All by RichardErkhov →

More Phi models

All Phi →

Other medium models

All medium models →