model card: remove DRY guidance, keep sequential regime, add torchvision dep
Browse files
README.md
CHANGED
|
@@ -71,6 +71,11 @@ paddleocr_vl`, `PaddleOCRVLForConditionalGeneration`) — **no `trust_remote_cod
|
|
| 71 |
needed**, but it requires **transformers ≥ 5.15** (native `paddleocr_vl` support)
|
| 72 |
or **vLLM ≥ 0.26**. Decode **greedy** (`temperature=0`).
|
| 73 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
> **⚠ Position regime — required for correct output.** This model was trained in the
|
| 75 |
> **sequential** image-token M-RoPE regime (1-D positions), *not* the default "grid"
|
| 76 |
> regime. If the processor emits a non-zero `mm_token_type_ids`, structured pages
|
|
|
|
| 71 |
needed**, but it requires **transformers ≥ 5.15** (native `paddleocr_vl` support)
|
| 72 |
or **vLLM ≥ 0.26**. Decode **greedy** (`temperature=0`).
|
| 73 |
|
| 74 |
+
```bash
|
| 75 |
+
pip install "transformers>=5.15" torch torchvision accelerate pillow
|
| 76 |
+
# torchvision is required by the PaddleOCR-VL image processor.
|
| 77 |
+
```
|
| 78 |
+
|
| 79 |
> **⚠ Position regime — required for correct output.** This model was trained in the
|
| 80 |
> **sequential** image-token M-RoPE regime (1-D positions), *not* the default "grid"
|
| 81 |
> regime. If the processor emits a non-zero `mm_token_type_ids`, structured pages
|