Skip to content

phixtral-2x2_8

A mixture-of-experts model built from two phi-2 models

Visit official site
Category AI models

What phixtral-2x2_8 is

The model is described as a mixture-of-experts text generation model assembled from two microsoft/phi-2 models.

What you can do with it

  • Generate text with a mixture-of-experts model
  • Run the model at 4-bit precision on a GPU
  • Evaluate the model on named benchmarks
  • Work in a Colab notebook

Example uses

  • Generate text in a deep learning research project
  • Use the model for model evaluation
  • Execute the model at 4-bit precision on a GPU

Model notes

  • The design is inspired by the Mixtral-8x7B-v0.1 architecture
  • A custom mergekit mixtral branch is used
  • AGIEval, GPT4All, TruthfulQA and Bigbench are named
  • The material states it outperforms each individual expert model

Pros & cons

✓ What we like

  • Runs at 4-bit precision on a free T4 GPU
  • AGIEval, GPT4All, TruthfulQA and Bigbench are named

! What to watch out for

  • The model uses a custom mergekit branch
  • The listed benchmarks are named evaluation sets

FAQ

How large is the model?

The material states 4.46B parameters and F16 tensor type.

Where can it run?

The material names a free T4 GPU in a Colab notebook at 4-bit precision.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review