{ "hf_id": "baidu/Unlimited-OCR", "meta": { "title": "Unlimited-OCR", "slug": "unlimited-ocr", "provider": "Ernie (Baidu)", "description": "Baidu's state-of-the-art document-parsing model with Reference Sliding Window Attention (R-SWA), optimized for full-page OCR and markdown generation.", "date_added": "2026-06-28", "date_updated": "2026-06-28", "tasks": [ "multimodal" ], "performance_headline": "One-shot long-horizon document parsing to markdown", "related_recipes": [ "deepseek-ai/DeepSeek-OCR" ] }, "recommended_command": { "hardware": "h200", "strategy": "single_node_tp", "variant": "default", "node_count": 1, "deploy_type": "single_node", "env": {}, "docker_image": "vllm/vllm-openai:unlimited-ocr", "command": "vllm serve baidu/Unlimited-OCR \\\n --trust-remote-code \\\n --logits_processors vllm.model_executor.models.unlimited_ocr:NGramPerReqLogitsProcessor \\\n --no-enable-prefix-caching \\\n --mm-processor-cache-gb 0 \\\n --tensor-parallel-size 1", "argv": [ "vllm", "serve", "baidu/Unlimited-OCR", "--trust-remote-code", "--logits_processors", "vllm.model_executor.models.unlimited_ocr:NGramPerReqLogitsProcessor", "--no-enable-prefix-caching", "--mm-processor-cache-gb", "0", "--tensor-parallel-size", "1" ], "docker_command": "docker run --gpus all \\\n --privileged --ipc=host -p 8000:8000 \\\n -v ~/.cache/huggingface:/root/.cache/huggingface \\\n vllm/vllm-openai:unlimited-ocr baidu/Unlimited-OCR \\\n --trust-remote-code \\\n --logits_processors vllm.model_executor.models.unlimited_ocr:NGramPerReqLogitsProcessor \\\n --no-enable-prefix-caching \\\n --mm-processor-cache-gb 0 \\\n --tensor-parallel-size 1", "docker_argv": [ "docker", "run", "--gpus", "all", "--privileged", "--ipc=host", "-p", "8000:8000", "-v", "~/.cache/huggingface:/root/.cache/huggingface", "vllm/vllm-openai:unlimited-ocr", "baidu/Unlimited-OCR", "--trust-remote-code", "--logits_processors", "vllm.model_executor.models.unlimited_ocr:NGramPerReqLogitsProcessor", "--no-enable-prefix-caching", "--mm-processor-cache-gb", "0", "--tensor-parallel-size", "1" ], "strategy_spec": { "name": "single_node_tp", "deploy_type": "single_node", "display_name": "Tensor Parallel", "short_label": "TP", "orientation": "latency", "description": "Single-node tensor parallel. Splits the model across all local GPUs. TP size is set to the GPU count at deploy time. The simplest multi-GPU strategy — works for all model architectures.\n", "hardware_match": { "min_gpus": 1, "max_gpus": 8, "multi_node": false }, "vllm_args": [], "parallel_flag": "--tensor-parallel-size" }, "hardware_profile": { "brand": "NVIDIA", "generation": "hopper", "compute_arch": "sm90", "display_name": "H200", "description": "NVIDIA H200 SXM 141 GB HBM3e · 8-GPU HGX node", "gpu_count": 8, "vram_gb": 1128, "multi_node": false }, "alternatives": { "multi_node_tp": "/baidu/Unlimited-OCR/strategies/multi_node_tp.json", "kv_store_centralized_mooncake": "/baidu/Unlimited-OCR/strategies/kv_store_centralized_mooncake.json", "kv_store_distributed_mooncake": "/baidu/Unlimited-OCR/strategies/kv_store_distributed_mooncake.json" }, "by_hardware": { "h200": "/baidu/Unlimited-OCR/hw/h200.json", "h100": "/baidu/Unlimited-OCR/hw/h100.json", "b200": "/baidu/Unlimited-OCR/hw/b200.json", "gb200": "/baidu/Unlimited-OCR/hw/gb200.json", "b300": "/baidu/Unlimited-OCR/hw/b300.json", "gb300": "/baidu/Unlimited-OCR/hw/gb300.json", "mi300x": "/baidu/Unlimited-OCR/hw/mi300x.json", "mi325x": "/baidu/Unlimited-OCR/hw/mi325x.json", "mi350x": "/baidu/Unlimited-OCR/hw/mi350x.json", "mi355x": "/baidu/Unlimited-OCR/hw/mi355x.json" } }, "model": { "model_id": "baidu/Unlimited-OCR", "min_vllm_version": "0.25.0", "docker_image": "vllm/vllm-openai:unlimited-ocr", "architecture": "dense", "parameter_count": "3B", "active_parameters": "3B", "context_length": 32768, "base_args": [ "--trust-remote-code", "--logits_processors", "vllm.model_executor.models.unlimited_ocr:NGramPerReqLogitsProcessor", "--no-enable-prefix-caching", "--mm-processor-cache-gb", "0" ], "base_env": {}, "install": { "pip": { "command": "uv venv\nsource .venv/bin/activate\nuv pip install -U vllm --torch-backend auto", "note": "Unlimited-OCR (UnlimitedOCRForCausalLM) is supported since vLLM v0.25.0." }, "docker": { "command": "docker pull vllm/vllm-openai:unlimited-ocr" } } }, "features": { "encoder_parallel": { "description": "Run the vision encoder in data-parallel mode — avoids TP comm overhead on the small encoder.", "args": [ "--mm-encoder-tp-mode", "data" ] } }, "opt_in_features": [ "encoder_parallel" ], "variants": { "default": { "precision": "bf16", "vram_minimum_gb": 8, "description": "Full precision BF16 (~3B params)" } }, "hardware_overrides": {}, "guide": "## Overview\n\n[Unlimited-OCR](https://huggingface.co/baidu/Unlimited-OCR) is Baidu's\nDeepSeek-OCR-lineage document-parsing model. It shares the DeepSeek-OCR\n*gundam* vision stack (SAM-ViT-B + CLIP-L DeepEncoder, `base_size=1024` /\n`image_size=640` / crop) and ships with the same n-gram logits processor for\nhigh-quality, repetition-free document → markdown generation.\n\n## Prerequisites\n\n- **Hardware**: a single GPU with >=8 GB VRAM is enough for BF16 inference.\n- **vLLM**: served from the dedicated release image (the architecture is not yet\n in a stable pip wheel). Pull the image shown in the Install → Docker tab.\n\n## Serving recipe (required)\n\nThe model has **no chat template** and is trained for a specific prompt/decode\nrecipe — without it the model returns empty output. The required pieces:\n\n- register the no-repeat-ngram logits processor on the server:\n `--logits_processors vllm.model_executor.models.unlimited_ocr:NGramPerReqLogitsProcessor`\n- prompt text must begin with a literal `` (e.g. `document parsing.`)\n- `skip_special_tokens=False`\n- pass the processor args per request: `ngram_size=35`, `window_size=128`\n (use `window_size=1024` for multi-page / PDF input)\n\n## Client Usage\n\n### Online OCR serving\n\n```bash\ndocker run --rm --gpus all --network host --ipc host \\\n vllm/vllm-openai:unlimited-ocr \\\n baidu/Unlimited-OCR \\\n --trust-remote-code \\\n --logits_processors vllm.model_executor.models.unlimited_ocr:NGramPerReqLogitsProcessor \\\n --no-enable-prefix-caching \\\n --mm-processor-cache-gb 0\n```\n\n```python\nimport time\nfrom openai import OpenAI\n\nclient = OpenAI(api_key=\"EMPTY\", base_url=\"http://localhost:8000/v1\", timeout=3600)\n\nmessages = [\n {\n \"role\": \"user\",\n \"content\": [\n {\"type\": \"text\", \"text\": \"document parsing.\"},\n {\"type\": \"image_url\", \"image_url\": {\"url\": \"https://huggingface.co/baidu/Unlimited-OCR/resolve/main/assets/baidu.png\"}},\n ],\n }\n]\n\nstart = time.time()\nresponse = client.chat.completions.create(\n model=\"baidu/Unlimited-OCR\",\n messages=messages,\n max_tokens=8192,\n temperature=0.0,\n extra_body={\n \"skip_special_tokens\": False,\n \"vllm_xargs\": {\"ngram_size\": 35, \"window_size\": 128},\n },\n)\nprint(f\"Response costs: {time.time() - start:.2f}s\")\nprint(f\"Generated text: {response.choices[0].message.content}\")\n```\n\nThe raw output carries `<|ref|>…<|/ref|>` / `<|det|>…<|/det|>` grounding tokens;\nunwrap the `<|ref|>` text and drop the `<|det|>` coordinate boxes to get clean markdown.\n\n## Troubleshooting / Configuration Tips\n\n- **The custom logits processor is required** — without it long documents loop on\n `<|det|>` coordinate tokens.\n- Empty output almost always means the prompt is missing the literal ``\n prefix or `skip_special_tokens` was left `True`.\n- OCR tasks don't benefit from prefix caching / image reuse, so prefix caching and\n the mm-processor cache are disabled in the recipe.\n- Single-image input uses gundam (crop) mode; multi-image requests automatically\n fall back to non-crop (base) mode — use `window_size=1024` for those.\n\n## References\n\n- [Unlimited-OCR on Hugging Face](https://huggingface.co/baidu/Unlimited-OCR)\n- [vLLM multimodal inputs guide](https://docs.vllm.ai/en/latest/features/multimodal_inputs.html#offline-inference)\n", "source": { "source": "vllm", "repo": "vllm-project/vllm", "ref": "main", "sha": "389ffebe555705c85337025d5df108d0384d365e", "path": "recipes/models/baidu/Unlimited-OCR.yaml", "edit_url": "https://github.com/vllm-project/vllm/edit/main/recipes/models/baidu/Unlimited-OCR.yaml", "overlays": [] } }