Model reference · open weights
trocr-small-korean is an open-weight language model from team-lucid. trocr-small-korean (FP32) weighs 109 MB; the smallest configuration that runs it is RTX 3060 12 GB.
For self-hosting, compare this build's requirements below with the GPU you plan to rent.
Summary of the team-lucid/trocr-small-korean model card, 2026-10-01
What it is
| Released by | team-lucid |
|---|---|
| Released | 2023-06-30 |
| Parameters | 55M |
| VRAM | 109 MB for the weights |
What it runs on
How much memory each request adds isn't estimated yet for this architecture. The weights need at least the cards below, plus room for the context.
| Card | Weights alone |
|---|---|
| RTX 3060 12 GB | fits |
| RTX 4060 Ti 16 GB | fits |
| RTX 3090 24 GB | fits |
| RTX 4090 24 GB | fits |
| RTX 5090 32 GB | fits |
| L40S 48 GB | fits |
| A100 80 GB | fits |
| H100 80 GB | fits |
| RTX PRO 6000 Blackwell 96 GB | fits |
| DGX Spark (GB10) 128 GB unified | fits |
| H200 141 GB | fits |
| B200 180 GB | fits |
From the model card
TrOCR은 Encoder-Decoder 모델로, 이미지 트랜스포머 인코더와 텍스트 트랜스포머 디코더로 이루어져 있습니다. 이미지 인코더는 DeiT 가중치로 초기화되었고, 텍스트 디코더는 자체적으로 학습한 RoBERTa 가중치로 초기화되었습니다.
이 연구는 구글의 TPU Research Cloud(TRC)를 통해 지원받은 Cloud TPU로 학습되었습니다.
import torch
from transformers import VisionEncoderDecoderModel
model = VisionEncoderDecoderModel.from_pretrained("team-lucid/trocr-small-korean")
pixel_values = torch.rand(1, 3, 384, 384)
generated_ids = model.generate(pixel_values)
해당 모델은 synthtiger로 합성된 6M개의 이미지로 학습되었습니다
| Hyperparameter | Small |
|---|---|
| Warmup Steps | 4,000 |
| Learning Rates | 1e-4 |
| Batch Size | 512 |
| Weight Decay | 0.01 |
| Max Steps | 500,000 |
| Learning Rate Decay | 0.1 |
| \(Adam\beta_1\) | 0.9 |
| \(Adam\beta_2\) | 0.98 |
Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.