# EfficientViT (MSRA)

**Category:** Efficient/Mobile · **Code:**
[`timm/models/efficientvit_msra.py`](https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/efficientvit_msra.py)
· **Architectures:** 6 · **Pretrained weights:** 6 (6 with ImageNet-1k results)

EfficientViT from Microsoft Research, a memory efficient vision transformer with cascaded group attention, which feeds
each attention head a different split of the features.

## Papers

* [EfficientViT: Memory Efficient Vision Transformer with Cascaded Group Attention](https://arxiv.org/abs/2305.07027)

## Usage

See the [Quickstart](../quickstart) for loading pretrained weights, image preprocessing and fine-tuning, and [Feature
Extraction](../feature_extraction) for using the models as backbones.

```py
>>> import timm
>>> model = timm.create_model('efficientvit_m5.r224_in1k', pretrained=True)
>>> data_config = timm.data.resolve_model_data_config(model)
```

## Model API support

| Feature | `EfficientVitMsra` |
|---|---|
| [`features_only=True` feature maps](../model_api#intermediate-features) | ✓ (strides 16, 32, 64) |
| [`forward_intermediates()`](../model_api#intermediate-features) | ✓ |
| [`prune_intermediate_layers()`](../model_api#intermediate-features) | ✓ |
| [Any input size, no changes needed](../model_api#input-size) | ✓ |
| [`set_input_size()`](../model_api#input-size) | ✓ |
| [`dynamic_img_size=True`](../model_api#input-size) | – |
| [`set_grad_checkpointing()`](../model_api#training-hooks) | ✓ |
| [Layer-wise LR decay (`group_matcher()`)](../model_api#training-hooks) | ✓ |
| [`reset_classifier()`](../model_api#classifier) | ✓ |

✓ supported, ✓ (square) works for square sizes only, ✗ not supported, – not needed (the model already accepts any input
size). Checked by running each feature on a randomly initialized model (`EfficientVitMsra` with `efficientvit_m0` at
224), other architectures of the same class are expected to behave the same. See the [Model API](../model_api) page for
details.

## Architectures

| Architecture | Params (M) | GMACs | Input size | Pretrained weights |
|---|---:|---:|---:|---:|
| `efficientvit_m0` | 2.33 | 0.08 | 224 | 1 |
| `efficientvit_m1` | 2.96 | 0.17 | 224 | 1 |
| `efficientvit_m2` | 4.17 | 0.20 | 224 | 1 |
| `efficientvit_m3` | 6.88 | 0.26 | 224 | 1 |
| `efficientvit_m4` | 8.78 | 0.30 | 224 | 1 |
| `efficientvit_m5` | 12.44 | 0.52 | 224 | 1 |

GMACs and parameter counts are for the default input size and classifier of each architecture, from the [benchmark
results](https://github.com/huggingface/pytorch-image-models/tree/main/results).

## Pretrained weights

Top-1 is ImageNet-1k validation accuracy from
[`results-imagenet.csv`](https://github.com/huggingface/pytorch-image-models/blob/main/results/results-imagenet.csv) at
the listed eval size, - if the weight has no ImageNet-1k result (e.g. pretraining or image-text weights). Classes is the
size of the pretrained classifier, none for weights without one.

| Weight | Input size | Top-1 (%) | Eval size | Classes | License | Hub |
|---|---:|---:|---:|---:|---|---|
| `efficientvit_m0.r224_in1k` | 224 | 63.304 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/efficientvit_m0.r224_in1k) |
| `efficientvit_m1.r224_in1k` | 224 | 68.340 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/efficientvit_m1.r224_in1k) |
| `efficientvit_m2.r224_in1k` | 224 | 70.796 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/efficientvit_m2.r224_in1k) |
| `efficientvit_m3.r224_in1k` | 224 | 73.394 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/efficientvit_m3.r224_in1k) |
| `efficientvit_m4.r224_in1k` | 224 | 74.344 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/efficientvit_m4.r224_in1k) |
| `efficientvit_m5.r224_in1k` | 224 | 77.092 | 224 | 1000 | mit | [timm](https://huggingface.co/timm/efficientvit_m5.r224_in1k) |