Rename dots-ocr-1.5 → dots-mocr (final release name)

- dots.mocr is the official final name for what was dots.ocr-1.5
- Model now on HF directly: rednote-hilab/dots.mocr (no ModelScope mirror needed)
- Added SVG prompt mode (--prompt-mode svg) with auto temperature/top_p
- SVG-optimized variant: rednote-hilab/dots.mocr-svg
- Updated README with new model entry and detailed docs
- Tested: OCR mode 3/3 on L4, SVG mode 3/3 on L4

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

README.md

@@ -7,7 +7,7 @@ tags: [uv-script, ocr, vision-language-model, document-processing, hf-jobs]
 
 > Part of [uv-scripts](https://huggingface.co/uv-scripts) - ready-to-run ML tools powered by UV and HuggingFace Jobs.
 
-15 OCR models from 0.9B to 8B parameters. Pick a model, point at your dataset, get markdown — no setup required.
+19 OCR scripts covering models from 0.9B to 8B parameters. Pick a model, point at your dataset, get markdown — no setup required.
 
 ## 🚀 Quick Start
 
@@ -43,7 +43,7 @@ That's it! The script will:
 | `dots-ocr.py` | [DoTS.ocr](https://huggingface.co/Tencent/DoTS.ocr) | 1.7B | vLLM | 100+ languages |
 | `firered-ocr.py` | [FireRed-OCR](https://huggingface.co/FireRedTeam/FireRed-OCR) | 2.1B | vLLM | Qwen3-VL fine-tune, Apache 2.0 |
 | `nanonets-ocr.py` | [Nanonets-OCR-s](https://huggingface.co/nanonets/Nanonets-OCR-s) | 2B | vLLM | LaTeX, tables, forms |
-| `dots-ocr-1.5.py` | [DoTS.ocr-1.5](https://huggingface.co/davanstrien/dots.ocr-1.5) | 3B | vLLM | 7 prompt modes, layout + bbox, 100+ languages |
+| `dots-mocr.py` | [dots.mocr](https://huggingface.co/rednote-hilab/dots.mocr) | 3B | vLLM | 8 prompt modes incl. SVG generation, layout + bbox, 100+ languages |
 | `nanonets-ocr2.py` | [Nanonets-OCR2-3B](https://huggingface.co/nanonets/Nanonets-OCR2-s) | 3B | vLLM | Next-gen, Qwen2.5-VL base |
 | `deepseek-ocr-vllm.py` | [DeepSeek-OCR](https://huggingface.co/deepseek-ai/DeepSeek-OCR) | 4B | vLLM | 5 resolution + 5 prompt modes |
 | `deepseek-ocr.py` | [DeepSeek-OCR](https://huggingface.co/deepseek-ai/DeepSeek-OCR) | 4B | Transformers | Same model, Transformers backend |
@@ -392,7 +392,46 @@ Advanced reasoning-based OCR using [numind/NuMarkdown-8B-Thinking](https://huggi
 - 🔍 **Multi-column Layouts** - Handles complex document structures
 - ✨ **Thinking Traces** - Optional inclusion of reasoning process with `--include-thinking`
 
-### DoTS.ocr (`dots-ocr.py`)
+### dots.mocr (`dots-mocr.py`) — SVG generation + SOTA OCR
+
+Advanced multilingual OCR and SVG generation using [rednote-hilab/dots.mocr](https://huggingface.co/rednote-hilab/dots.mocr) with 3B parameters:
+
+- 🌍 **100+ Languages** - Extensive multilingual support
+- 📝 **Document OCR** - Clean text extraction (default mode)
+- 📊 **Layout Analysis** - Structured output with bboxes and categories
+- 📐 **Formula recognition** - LaTeX format support
+- 🖼️ **SVG generation** - Convert charts, UI layouts, figures to editable SVG code
+- 🔀 **8 prompt modes** - OCR, layout-all, layout-only, web-parsing, scene-spotting, grounding-ocr, svg, general
+- 📄 **[Paper](https://arxiv.org/abs/2603.13032)** - 83.9% on olmOCR-Bench
+
+**SVG variant:** Use `--model rednote-hilab/dots.mocr-svg` with `--prompt-mode svg` for best SVG results.
+
+**Quick start:**
+
+```bash
+# Basic OCR
+hf jobs uv run --flavor l4x1 \
+    -s HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/dots-mocr.py \
+    your-input-dataset your-output-dataset \
+    --max-samples 100
+
+# SVG generation from charts/figures
+hf jobs uv run --flavor l4x1 \
+    -s HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/dots-mocr.py \
+    your-charts svg-output \
+    --prompt-mode svg --model rednote-hilab/dots.mocr-svg
+
+# Layout analysis with bounding boxes
+hf jobs uv run --flavor l4x1 \
+    -s HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/dots-mocr.py \
+    your-documents layout-output \
+    --prompt-mode layout-all
+```
+
+### DoTS.ocr v1 (`dots-ocr.py`)
 
 Compact multilingual OCR using [rednote-hilab/dots.ocr](https://huggingface.co/rednote-hilab/dots.ocr) with only 1.7B parameters:
 

dots-ocr-1.5.py → dots-mocr.py

@@ -13,23 +13,27 @@
 # ///
 
 """
-Convert document images to markdown using DoTS.ocr-1.5 with vLLM.
+Convert document images to markdown using dots.mocr with vLLM.
 
-DoTS.ocr-1.5 is a 3B multilingual document parsing model with SOTA performance
-on 100+ languages. Compared to v1 (1.7B), it adds web screen parsing, scene text
-spotting, SVG code generation, and stronger multilingual document parsing.
+dots.mocr is a 3B multilingual document parsing model with SOTA performance
+on 100+ languages. It excels at converting structured graphics (charts, UI
+layouts, scientific figures) directly into SVG code. Core capabilities include
+grounding, recognition, semantic understanding, and interactive dialogue.
 
 Features:
 - Multilingual support (100+ languages)
 - Table extraction and formatting
 - Formula recognition
 - Layout-aware text extraction
-- Web screen parsing (NEW in v1.5)
-- Scene text spotting (NEW in v1.5)
-- SVG code generation (requires dots.ocr-1.5-svg variant)
-
-Model: rednote-hilab/dots.ocr-1.5
-vLLM: Officially supported (same DotsOCRForCausalLM architecture as v1)
+- Web screen parsing
+- Scene text spotting
+- SVG code generation (use --prompt-mode svg, or --model rednote-hilab/dots.mocr-svg for best results)
+
+Model: rednote-hilab/dots.mocr
+SVG variant: rednote-hilab/dots.mocr-svg
+vLLM: Officially integrated since v0.11.0
+GitHub: https://github.com/rednote-hilab/dots.mocr
+Paper: https://arxiv.org/abs/2603.13032
 """
 
 import argparse
@@ -56,8 +60,8 @@ logger = logging.getLogger(__name__)
 
 
 # ────────────────────────────────────────────────────────────────
-# DoTS OCR 1.5 Prompt Templates (from official dots.ocr repo)
-# Source: https://github.com/rednote-hilab/dots.ocr/blob/master/dots_ocr/utils/prompts.py
+# dots.mocr Prompt Templates
+# Source: https://github.com/rednote-hilab/dots.mocr/blob/master/dots_mocr/utils/prompts.py
 # ────────────────────────────────────────────────────────────────
 
 PROMPT_TEMPLATES = {
@@ -86,10 +90,13 @@ PROMPT_TEMPLATES = {
     #   resized_h, resized_w = smart_resize(orig_h, orig_w)
     #   scale_x, scale_y = orig_w / resized_w, orig_h / resized_h
     "layout-only": """Please output the layout information from this PDF image, including each layout's bbox and its category. The bbox should be in the format [x1, y1, x2, y2]. The layout categories for the PDF document include ['Caption', 'Footnote', 'Formula', 'List-item', 'Page-footer', 'Page-header', 'Picture', 'Section-header', 'Table', 'Text', 'Title']. Do not output the corresponding text. The layout result should be in JSON format.""",
-    # NEW in v1.5:
     "web-parsing": """Parsing the layout info of this webpage image with format json:\n""",
     "scene-spotting": """Detect and recognize the text in the image.""",
     "grounding-ocr": """Extract text from the given bounding box on the image (format: [x1, y1, x2, y2]).\nBounding Box:\n""",
+    # SVG code generation — {width} and {height} are replaced with actual image dimensions.
+    # For best results, use --model rednote-hilab/dots.mocr-svg
+    # Uses higher temperature (0.9) and top_p (1.0) per official recommendation.
+    "svg": """Please generate the SVG code based on the image. viewBox="0 0 {width} {height}" """,
     "general": """ """,
 }
 
@@ -122,6 +129,12 @@ def make_ocr_message(
     # Convert to RGB
     pil_img = pil_img.convert("RGB")
 
+    # For SVG mode, inject actual image dimensions into the prompt
+    if "{width}" in prompt and "{height}" in prompt:
+        prompt = prompt.replace("{width}", str(pil_img.width)).replace(
+            "{height}", str(pil_img.height)
+        )
+
     # Convert to base64 data URI
     buf = io.BytesIO()
     pil_img.save(buf, format="PNG")
@@ -159,7 +172,7 @@ def create_dataset_card(
 tags:
 - ocr
 - document-processing
-- dots-ocr-1.5
+- dots-mocr
 - multilingual
 - markdown
 - uv-script
@@ -168,7 +181,7 @@ tags:
 
 # Document OCR using {model_name}
 
-This dataset contains OCR results from images in [{source_dataset}](https://huggingface.co/datasets/{source_dataset}) using DoTS.ocr-1.5, a 3B multilingual model with SOTA document parsing.
+This dataset contains OCR results from images in [{source_dataset}](https://huggingface.co/datasets/{source_dataset}) using dots.mocr, a 3B multilingual model with SOTA document parsing and SVG generation.
 
 ## Processing Details
 
@@ -191,13 +204,14 @@ This dataset contains OCR results from images in [{source_dataset}](https://hugg
 
 ## Model Information
 
-DoTS.ocr-1.5 is a 3B multilingual document parsing model that excels at:
+dots.mocr is a 3B multilingual document parsing model that excels at:
 - 100+ Languages — Multilingual document support
 - Table extraction — Structured data recognition
 - Formulas — Mathematical notation preservation
 - Layout-aware — Reading order and structure preservation
 - Web screen parsing — Webpage layout analysis
 - Scene text spotting — Text detection in natural scenes
+- SVG code generation — Charts, UI layouts, scientific figures to SVG
 
 ## Dataset Structure
 
@@ -227,10 +241,10 @@ for info in inference_info:
 
 ## Reproduction
 
-This dataset was generated using the [uv-scripts/ocr](https://huggingface.co/datasets/uv-scripts/ocr) DoTS OCR 1.5 script:
+This dataset was generated using the [uv-scripts/ocr](https://huggingface.co/datasets/uv-scripts/ocr) dots.mocr script:
 
 ```bash
-uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/dots-ocr-1.5.py \\
+uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/dots-mocr.py \\
     {source_dataset} \\
     <output-dataset> \\
     --image-column {image_column} \\
@@ -250,7 +264,7 @@ def main(
     output_dataset: str,
     image_column: str = "image",
     batch_size: int = 16,
-    model: str = "rednote-hilab/dots.ocr-1.5",
+    model: str = "rednote-hilab/dots.mocr",
     max_model_len: int = 24000,
     max_tokens: int = 24000,
     gpu_memory_utilization: float = 0.9,
@@ -269,7 +283,7 @@ def main(
     top_p: float = 0.9,
     verbose: bool = False,
 ):
-    """Process images from HF dataset through DoTS.ocr-1.5 model."""
+    """Process images from HF dataset through dots.mocr model."""
 
     # Check CUDA availability first
     check_cuda_availability()
@@ -320,6 +334,12 @@ def main(
         gpu_memory_utilization=gpu_memory_utilization,
     )
 
+    # SVG mode uses higher temperature/top_p per official recommendation
+    if prompt_mode == "svg" and temperature == 0.1 and top_p == 0.9:
+        logger.info("SVG mode: using recommended temperature=0.9, top_p=1.0")
+        temperature = 0.9
+        top_p = 1.0
+
     sampling_params = SamplingParams(
         temperature=temperature,
         top_p=top_p,
@@ -335,7 +355,7 @@ def main(
     for batch_indices in tqdm(
         partition_all(batch_size, range(len(dataset))),
         total=(len(dataset) + batch_size - 1) // batch_size,
-        desc="DoTS.ocr-1.5 processing",
+        desc="dots.mocr processing",
     ):
         batch_indices = list(batch_indices)
         batch_images = [dataset[i][image_column] for i in batch_indices]
@@ -344,7 +364,7 @@ def main(
             # Create messages for batch
             batch_messages = [make_ocr_message(img, prompt) for img in batch_images]
 
-            # Process with vLLM (dots.ocr-1.5 needs "string" content format)
+            # Process with vLLM (dots.mocr needs "string" content format)
             outputs = llm.chat(
                 batch_messages,
                 sampling_params,
@@ -372,7 +392,7 @@ def main(
     # Handle inference_info tracking (for multi-model comparisons)
     inference_entry = {
         "model_id": model,
-        "model_name": "DoTS.ocr-1.5",
+        "model_name": "dots.mocr",
         "column_name": output_column,
         "timestamp": datetime.now().isoformat(),
         "prompt_mode": prompt_mode if not custom_prompt else "custom",
@@ -453,7 +473,7 @@ def main(
     card = DatasetCard(card_content)
     card.push_to_hub(output_dataset, token=HF_TOKEN)
 
-    logger.info("DoTS.ocr-1.5 processing complete!")
+    logger.info("dots.mocr processing complete!")
     logger.info(
         f"Dataset available at: https://huggingface.co/datasets/{output_dataset}"
     )
@@ -475,77 +495,83 @@ if __name__ == "__main__":
     # Show example usage if no arguments
     if len(sys.argv) == 1:
         print("=" * 80)
-        print("DoTS.ocr-1.5 Document Processing")
+        print("dots.mocr Document Processing")
         print("=" * 80)
-        print("\n3B multilingual OCR model supporting 100+ languages")
+        print("\n3B multilingual OCR model with SVG generation")
         print("\nFeatures:")
         print("- Multilingual support (100+ languages)")
         print("- Fast processing with vLLM")
         print("- Table extraction and formatting")
         print("- Formula recognition")
         print("- Layout-aware text extraction")
-        print("- Web screen parsing (NEW in v1.5)")
-        print("- Scene text spotting (NEW in v1.5)")
+        print("- Web screen parsing")
+        print("- Scene text spotting")
+        print("- SVG code generation (charts, UI, figures)")
         print("\nPrompt modes:")
-        print("  ocr           - Text extraction (default)")
-        print("  layout-all    - Layout + bboxes + text (JSON)")
-        print("  layout-only   - Layout + bboxes only (JSON)")
-        print("  web-parsing   - Webpage layout analysis (JSON)")
+        print("  ocr            - Text extraction (default)")
+        print("  layout-all     - Layout + bboxes + text (JSON)")
+        print("  layout-only    - Layout + bboxes only (JSON)")
+        print("  web-parsing    - Webpage layout analysis (JSON)")
         print("  scene-spotting - Scene text detection")
-        print("  grounding-ocr - Text from bounding box region")
-        print("  general       - Free-form (use with --custom-prompt)")
+        print("  grounding-ocr  - Text from bounding box region")
+        print("  svg            - SVG code generation")
+        print("  general        - Free-form (use with --custom-prompt)")
         print("\nExample usage:")
         print("\n1. Basic OCR:")
-        print("   uv run dots-ocr-1.5.py input-dataset output-dataset")
-        print("\n2. Web screen parsing:")
-        print("   uv run dots-ocr-1.5.py screenshots parsed --prompt-mode web-parsing")
-        print("\n3. Scene text spotting:")
-        print("   uv run dots-ocr-1.5.py photos detected --prompt-mode scene-spotting")
+        print("   uv run dots-mocr.py input-dataset output-dataset")
+        print("\n2. SVG generation:")
+        print(
+            "   uv run dots-mocr.py charts svg-output --prompt-mode svg --model rednote-hilab/dots.mocr-svg"
+        )
+        print("\n3. Web screen parsing:")
+        print("   uv run dots-mocr.py screenshots parsed --prompt-mode web-parsing")
         print("\n4. Layout analysis with structure:")
-        print("   uv run dots-ocr-1.5.py papers analyzed --prompt-mode layout-all")
+        print("   uv run dots-mocr.py papers analyzed --prompt-mode layout-all")
         print("\n5. Running on HF Jobs:")
         print("   hf jobs uv run --flavor l4x1 \\")
         print("     -s HF_TOKEN \\")
         print(
-            "     https://huggingface.co/datasets/uv-scripts/ocr/raw/main/dots-ocr-1.5.py \\"
+            "     https://huggingface.co/datasets/uv-scripts/ocr/raw/main/dots-mocr.py \\"
         )
         print("       input-dataset output-dataset")
         print("\n" + "=" * 80)
-        print("\nFor full help, run: uv run dots-ocr-1.5.py --help")
+        print("\nFor full help, run: uv run dots-mocr.py --help")
         sys.exit(0)
 
     parser = argparse.ArgumentParser(
-        description="Document OCR using DoTS.ocr-1.5 (3B multilingual model)",
+        description="Document OCR using dots.mocr (3B multilingual model with SVG generation)",
         formatter_class=argparse.RawDescriptionHelpFormatter,
         epilog="""
-Prompt Modes (official DoTS.ocr-1.5 prompts):
+Prompt Modes (official dots.mocr prompts):
   ocr            - Simple text extraction (default)
   layout-all     - Layout analysis with bboxes, categories, and text (JSON output)
   layout-only    - Layout detection with bboxes and categories only (JSON output)
-  web-parsing    - Webpage layout analysis (JSON output) [NEW in v1.5]
-  scene-spotting - Scene text detection and recognition [NEW in v1.5]
-  grounding-ocr  - Extract text from bounding box region [NEW in v1.5]
-  general        - Free-form QA (use with --custom-prompt) [NEW in v1.5]
+  web-parsing    - Webpage layout analysis (JSON output)
+  scene-spotting - Scene text detection and recognition
+  grounding-ocr  - Extract text from bounding box region
+  svg            - SVG code generation (auto-injects image dimensions into viewBox)
+  general        - Free-form QA (use with --custom-prompt)
 
 SVG Code Generation:
-  For SVG output, use --model rednote-hilab/dots.ocr-1.5-svg with:
-  --custom-prompt 'Please generate the SVG code based on the image.'
+  Use --prompt-mode svg for SVG output. For best results, combine with
+  --model rednote-hilab/dots.mocr-svg (the SVG-optimized variant).
+  SVG mode automatically uses temperature=0.9, top_p=1.0 unless overridden.
 
 Examples:
   # Basic text OCR (default)
-  uv run dots-ocr-1.5.py my-docs analyzed-docs
+  uv run dots-mocr.py my-docs analyzed-docs
 
-  # Web screen parsing
-  uv run dots-ocr-1.5.py screenshots parsed --prompt-mode web-parsing
+  # SVG generation with optimized variant
+  uv run dots-mocr.py charts svg-out --prompt-mode svg --model rednote-hilab/dots.mocr-svg
 
-  # Scene text spotting
-  uv run dots-ocr-1.5.py photos spotted --prompt-mode scene-spotting
+  # Web screen parsing
+  uv run dots-mocr.py screenshots parsed --prompt-mode web-parsing
 
   # Full layout analysis with structure
-  uv run dots-ocr-1.5.py papers structured --prompt-mode layout-all
+  uv run dots-mocr.py papers structured --prompt-mode layout-all
 
   # Random sampling for testing
-  uv run dots-ocr-1.5.py large-dataset test --max-samples 50 --shuffle
+  uv run dots-mocr.py large-dataset test --max-samples 50 --shuffle
         """,
     )
 
@@ -564,8 +590,8 @@ Examples:
     )
     parser.add_argument(
         "--model",
-        default="rednote-hilab/dots.ocr-1.5",
-        help="Model to use (default: rednote-hilab/dots.ocr-1.5)",
+        default="rednote-hilab/dots.mocr",
+        help="Model to use (default: rednote-hilab/dots.mocr, or rednote-hilab/dots.mocr-svg for SVG)",
     )
     parser.add_argument(
         "--max-model-len",