MiniCPM-OCR
MiniCPM-V Multimodal Series
MiniCPM-OCR 1.0
Choose weights for your inference backend. A quantized checkpoint still needs a backend version that supports the model architecture.
Recognition weights
| Format | HuggingFace | ModelScope |
|---|---|---|
| BF16 | MiniCPM-OCR-1.0 | MiniCPM-OCR-1.0 |
| GGUF | MiniCPM-OCR-1.0-gguf | MiniCPM-OCR-1.0-gguf |
| AWQ | MiniCPM-OCR-1.0-AWQ | MiniCPM-OCR-1.0-AWQ |
| GPTQ | MiniCPM-OCR-1.0-GPTQ | MiniCPM-OCR-1.0-GPTQ |
DSpark
DSpark is speculative decoding acceleration and uses a separate draft model. Download a draft matching the recognition model release and backend format.
| Format | HuggingFace | ModelScope |
|---|---|---|
| BF16 | MiniCPM-OCR-1.0-DSpark | MiniCPM-OCR-1.0-DSpark |
| GGUF | MiniCPM-OCR-1.0-DSpark-gguf | MiniCPM-OCR-1.0-DSpark-gguf |
See Deployment for the BF16 + DSpark service command. For GGUF, use the llama.cpp guide. Measure accuracy and throughput on your documents after switching formats.
MiniCPM-OCR Jev
Model downloads
| Format | HuggingFace | ModelScope |
|---|---|---|
| BF16 | MiniCPM-OCR-Jev | MiniCPM-OCR-Jev |
| GGUF | MiniCPM-OCR-Jev-gguf | MiniCPM-OCR-Jev-gguf |
| AWQ | MiniCPM-OCR-Jev-AWQ | MiniCPM-OCR-Jev-AWQ |
| GPTQ | MiniCPM-OCR-Jev-GPTQ | MiniCPM-OCR-Jev-GPTQ |