Visual Question Answering on DocVQA (held-out)
82.8ANLSGPT-4+OCR
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-4+OCRSize=~1T, Setting=ZS2023.12 | 82.8 | |
| DocLLM-7BSize=7B, Setting=STDD2023.12 | 63.4 | |
| DocLLM-1BSize=1B, Setting=STDD2023.12 | 53.5 | |
| Llama2+OCRSize=7B, Setting=ZS2023.12 | 47.4 |