Image-Text-to-Text
MLX
Safetensors
mistral3
ocr
document-understanding
vision-language
pdf
tables
forms
conversational
4-bit precision
Instructions to use humpf/LightOnOCR-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use humpf/LightOnOCR-4bit with MLX:
# Make sure mlx-vlm is installed # pip install --upgrade mlx-vlm from mlx_vlm import load, generate from mlx_vlm.prompt_utils import apply_chat_template from mlx_vlm.utils import load_config # Load the model model, processor = load("humpf/LightOnOCR-4bit") config = load_config("humpf/LightOnOCR-4bit") # Prepare input image = ["http://images.cocodataset.org/val2017/000000039769.jpg"] prompt = "Describe this image." # Apply chat template formatted_prompt = apply_chat_template( processor, config, prompt, num_images=1 ) # Generate output output = generate(model, processor, formatted_prompt, image) print(output) - Notebooks
- Google Colab
- Kaggle
- Local Apps
- LM Studio
| { | |
| "architectures": [ | |
| "LightOnOCRForConditionalGeneration" | |
| ], | |
| "dtype": "bfloat16", | |
| "eos_token_id": [ | |
| 151645, | |
| 151643 | |
| ], | |
| "image_token_id": 151655, | |
| "model_type": "mistral3", | |
| "multimodal_projector_bias": false, | |
| "pad_token_id": 151643, | |
| "projector_hidden_act": "gelu", | |
| "quantization": { | |
| "group_size": 64, | |
| "bits": 4, | |
| "mode": "affine" | |
| }, | |
| "quantization_config": { | |
| "group_size": 64, | |
| "bits": 4, | |
| "mode": "affine" | |
| }, | |
| "spatial_merge_size": 2, | |
| "text_config": { | |
| "architectures": [ | |
| "Qwen3ForCausalLM" | |
| ], | |
| "attention_bias": false, | |
| "attention_dropout": 0, | |
| "dtype": "bfloat16", | |
| "head_dim": 128, | |
| "hidden_act": "silu", | |
| "hidden_size": 1024, | |
| "initializer_range": 0.02, | |
| "intermediate_size": 3072, | |
| "layer_types": [ | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention", | |
| "full_attention" | |
| ], | |
| "max_position_embeddings": 16384, | |
| "max_window_layers": 28, | |
| "model_type": "qwen3", | |
| "num_attention_heads": 16, | |
| "num_hidden_layers": 28, | |
| "num_key_value_heads": 8, | |
| "rms_norm_eps": 1e-06, | |
| "rope_parameters": { | |
| "rope_theta": 1000000, | |
| "rope_type": "default" | |
| }, | |
| "rope_theta": 1000000, | |
| "sliding_window": null, | |
| "use_cache": true, | |
| "use_sliding_window": false, | |
| "tie_word_embeddings": true, | |
| "vocab_size": 151936, | |
| "use_qk_norm": true | |
| }, | |
| "transformers_version": "5.0.0.dev0", | |
| "use_cache": false, | |
| "vision_config": { | |
| "attention_dropout": 0, | |
| "dtype": "bfloat16", | |
| "head_dim": 64, | |
| "hidden_act": "silu", | |
| "hidden_size": 1024, | |
| "image_size": 1540, | |
| "initializer_range": 0.02, | |
| "intermediate_size": 4096, | |
| "model_type": "pixtral", | |
| "num_attention_heads": 16, | |
| "num_channels": 3, | |
| "num_hidden_layers": 24, | |
| "patch_size": 14, | |
| "rope_parameters": { | |
| "rope_theta": 10000, | |
| "rope_type": "default" | |
| }, | |
| "rope_theta": 10000 | |
| }, | |
| "vision_feature_layer": -1 | |
| } |