---
title: Phi4-multimodal-instruct
canonical_url: "https://www.modelscope.ai/models/AMD/Phi4-multimodal-instruct"
md_url: "https://www.modelscope.ai/models/AMD/Phi4-multimodal-instruct.md"
repository: AMD/Phi4-multimodal-instruct
last_updated: 2026-11-11
model_type:
  - phi4mm
architectures:
  - Phi4MMForCausalLM
parameters: 5.6B
tensor_type:
  - BF16
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 516
stars: 0
---

# Phi4-multimodal-instruct

> Phi4-multimodal-instruct - An open-source model by AMD on ModelScope. This model is a clone of microsoft/Phi-4-multimodal-instruct model. The changes in this repo, as compared to the original, has an optimized version of RoPE cached Embedding layer. This…

- **Repository**: AMD/Phi4-multimodal-instruct
- **Parameters**: 5.6B
- **Downloads**: 516
- **Stars**: 0
- **Last updated**: 2026-11-11

Source: https://www.modelscope.ai/models/AMD/Phi4-multimodal-instruct

---

This model is a clone of `microsoft/Phi-4-multimodal-instruct` model. The changes in this repo, as compared to the original, has an optimized version of RoPE cached Embedding layer. This change can be found Phi4MMLongRoPEScaledRotaryEmbedding class in `modeling_phi4mm.py` file.