deepseek-ai

/

DeepSeek-OCR

like 3.35k

Follow

DeepSeek 143k

Pretty plz

This would be very helpful. I hope someone is working on it.

+1

使用 llama-cpp 可以自己转换吧

GGUF Available in NexaSDK (only). Here's how to run it locally:

Model Card: https://huggingface.co/NexaAI/DeepSeek-OCR-GGUF

Quickstart

1. Install NexaSDK

2. Run the model locally with one line of code:

   nexa infer NexaAI/DeepSeek-OCR-GGUF
   

3. Then drag your image to terminal or type into the image path

case 1 : extract text

<your-image-path> Free OCR.

case 2 : extract bounding box

<your-image-path> <|grounding|>Convert the document to markdown. 

Note: If the model fails to run, install the latest Vulkan driver for Windows

There is a model that runs directly in ollama now.

You will need the latest docker image:
ollama/ollama:0.13.0-rc0 (0.12.11 will not work)

you use the following image:
ollama run mike/deepseek-ocr

Enjoy

hey @jcuypers , unlike from their official page, how we can use the model parameters in the inference call? for example

model.infer(
        tokenizer,
        prompt=prompt,
        image_file=image_path,
        output_path=output_dir,
        base_size=xxxx,
        image_size=xxxx,
        crop_mode=xxxxx,
        save_results=xxxx,
        test_compress=xxxx
    )

hi @gsvc , haven't tried it yet, but i guess it would have to be through the kwargs arguments used in some other tools (like langchain -> langchain_ollama / langchain_openai). similar as temperature etc is set through these frameworks. maybe someone else has tried this and can comment. thanks.

ok @jcuypers . But from the ollama documentation page its only have the temperature as parameter.

@gsvc , it is possible to use others. I personally used "num_ctx" and it loaded the model with the provided context size. Also openwebui allows for customization directly to ollama, so there is a possibility. not sure whether it was through ollama or openai interface, but it is available. this doesn't solve your problem since the list of parameters is fixed.

I think what you are looking for is here:
https://reference.langchain.com/python/integrations/langchain_openai/ChatOpenAI/#langchain_openai.chat_models.ChatOpenAI
model_kwargs / extra_body section

gl,j.