| | --- |
| | license: mit |
| | library_name: rkllm |
| | base_model: deepseek-ai/DeepSeek-OCR |
| | tags: |
| | - rkllm |
| | - rk3588 |
| | - npu |
| | - rockchip |
| | - quantized |
| | - edge-ai |
| | - orange-pi |
| | - vision-language |
| | - ocr |
| | model_name: DeepSeek-OCR-RKLLM |
| | pipeline_tag: image-text-to-text |
| | language: |
| | - en |
| | --- |
| | |
| |
|
| | # DeepSeek-OCR for RK3588 |
| |
|
| | This repository contains the optimized **RKNN** (Vision) and **RKLLM** (Language) models for running **DeepSeek-OCR** on Rockchip RK3588 NPUs. |
| |
|
| | ### 📦 Files |
| |
|
| | | File | Type | Description | |
| | | --- | --- | --- | |
| | | `deepseekocr_vision_rk3588.rknn` | **RKNN** | Vision Encoder (ViT) | |
| | | `deepseekocr_w8a8_rk3588.rkllm` | **RKLLM** | Language Decoder (w8a8) | |
| |
|
| | --- |
| |
|
| | ### ⚙️ System Requirements |
| |
|
| | * **NPU Driver:** `≥ 0.9.8` |
| | * **RKLLM Runtime:** `v1.2.3` |
| | * **RKNN-Toolkit2:** `v2.3.0` |
| |
|
| | ### 💻 Tested Hardware |
| |
|
| | * **Device:** Orange Pi 5 Pro |
| | * **RAM:** 16GB |
| | * **OS:** Armbian OS |
| |
|
| | --- |
| |
|
| | ### 🙏 Acknowledgements |
| |
|
| | * [DeepSeek Team](https://huggingface.co/deepseek-ai) for the base model architecture. |
| | * [Rockchip / airockchip](https://github.com/airockchip/rknn-llm) for the RKLLM toolkit and runtime. |
| | * Converted by [Akmal Dira](https://huggingface.co/akmaldira). |
| |
|
| | --- |
| |
|
