Distilled base DeiT at 224x224
According to the model card, this is the distilled base-sized DeiT variant, pre-trained and fine-tuned with distillation on ImageNet-1k at 224x224.
Open Source Model Profile · facebook
deit-base-distilled-patch16-224
deit-base-distilled-patch16-224 is a distilled base-sized DeiT image-classification model from facebook. According to the model card, it was pre-trained and fine-tuned with distillation on ImageNet-1k at 224x224 resolution.
- Publisher
- Task
- image-classification
- Model type
- deit
- License
- apache-2.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
deit-base-distilled-patch16-224 is published by facebook as a distilled Data-efficient Image Transformer for image classification. The captured configuration identifies DeiTForImageClassificationWithTeacher with model type deit. According to the model card, it was pre-trained and fine-tuned on ImageNet-1k (1 million images, 1,000 classes) at 224x224, with weights converted from the timm repository.
Recorded capabilities
Distillation token design
According to the model card, the model uses a distillation token besides the class token to learn from a teacher CNN during pre-training and fine-tuning.
16x16 patch input
According to the model card, images are presented as a sequence of fixed-size 16x16 patches which are linearly embedded.
Publisher-reported 83.4% top-1
According to the model card's results table, the DeiT-base distilled entry reports 83.4 ImageNet top-1 and 96.5 top-5 accuracy at 87M parameters.
Use cases in the source record
- Image classification of inputs into one of the 1,000 ImageNet classes using the model card's documented Transformers workflow.
- Distillation-technique study using the model card's description of the distillation token learning from a CNN teacher alongside class and patch tokens.
Limitations and unknowns
- No Safetensors parameter count was extracted from this record; the only size figure is the model card table's 87M entry.
- No independently measured evaluation results were extracted; accuracy figures are publisher-reported card claims.
- According to the card disclaimer, the releasing team did not write the model card; the Hugging Face team wrote it instead.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: facebook/deit-base-distilled-patch16-224
Captured: Unknown. Processed: 2026-09-07T19:34:44.365390+00:00.
Distilled Data-efficient Image Transformer (base-sized model) Distilled data-efficient Image Transformer (DeiT) model pre-trained and fine-tuned on ImageNet-1k (1 million images, 1,000 classes) at resolution 224x224. It was first introduced in the paper Training data-efficient image transformers & distillation through attention by Touvron et al. and first released in this repository . However, the weights were converted from the timm repository by Ross Wightman. Disclaimer: The team releasing DeiT did not write a model card for this model so this model card has been written by the Hugging Face team. Model description This model is a…
F001F002F003F004F005F006F009F010F011F012F013F014F015F019