Using Docling Offline

[egesnr]

Hello,
I'm trying to use docling offline due to api restrictions. Can i download the model using the jupyter terminal ? If I also can't download the hugging face manuelly in this pc what should i do?

[dosubot[bot]]

Yes, you can use Docling offline if all required models are available locally. To download models using the Jupyter terminal, run:

docling-tools models download --all -o models

This command downloads all supported models into a local folder named models. Make sure Docling is configured to load models from this folder by setting the DOCLING_SERVE_ARTIFACTS_PATH environment variable to the path where your models are stored.

If you can't download models directly from Hugging Face or the internet on your PC, you can download them on another machine, transfer the models folder to your PC (for example, using a USB drive or network share), and then set DOCLING_SERVE_ARTIFACTS_PATH to point to the transferred folder. Docling will use the models from that location and won't attempt to download anything at runtime. If a required model is missing, Docling will raise a runtime error rather than downloading it automatically. EasyOCR models are included by default in recent versions, so you don't need to download them separately unless you want to customize your model set.

More details and examples are available in the Docling Knowledge Base.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[egesnr]

@dosu when i try to run offline after downloaded models to local path and using the code you sent, i get this: raise LocalEntryNotFoundError(
huggingface_hub.errors.LocalEntryNotFoundError: An error happened while trying to locate the files on the Hub and we cannot find the appropriate snapshot folder for the specified revision on the local disk. Please check your internet connection and try again.
-> Cannot close object, library is destroyed. This may cause a memory leak!

[dosubot[bot]]

You've reached the anonymous usage limit of 10 interactions per month. Your usage will reset on 2026-01-01. If you'd like to use Dosu before then, please register to increase your limit to 100 interactions per month for free.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[egesnr]

@dosu raise LocalEntryNotFoundError(
huggingface_hub.errors.LocalEntryNotFoundError: An error happened while trying to locate the files on the Hub and we cannot find the appropriate snapshot folder for the specified revision on the local disk. Please check your internet connection and try again.
-> Cannot close object, library is destroyed. This may cause a memory leak! why i get this error why the code cant find my locally downloaded models i exactly used the code you sent

[dosubot[bot]]

This error means Docling can't find your local models because the folder structure or environment variable isn't exactly what it expects. Here’s what to check:

1. The environment variable must be DOCLING_SERVE_ARTIFACTS_PATH (not DOCLING_ARTIFACTS_PATH), and it must point to the parent directory containing all model subfolders—not to a specific model folder or to the Hugging Face cache root. Use an absolute path, not a relative one. Set it before starting Python, or at the very top of your script, before importing Docling code (reference).

2. Each model must be in its own subfolder, named by replacing / with -- in the repo ID. For current Docling versions, the folder names must start with docling-project--, not ds4sd--. For example:

   /your_models_dir/
     docling-project--docling-layout-heron/
       model.safetensors
       config.json
       preprocessor_config.json
     docling-project--docling-models/
       model_artifacts/
         tableformer/
           model.safetensors
           config.json
           preprocessor_config.json
   

   If your folders use the old ds4sd-- prefix, rename them to docling-project--... (reference).

3. All required files (model.safetensors, config.json, preprocessor_config.json) must be present and directly inside each model’s subfolder (not nested too deeply). For table structure, files go under model_artifacts/tableformer/ inside docling-project--docling-models (reference).

4. In your code, explicitly pass the absolute path to artifacts_path in PdfPipelineOptions:

   pipeline_options = PdfPipelineOptions(
       artifacts_path="/absolute/path/to/your_models_dir",
       do_ocr=False
   )

   This is required for Docling v2.60.0+ (reference).

5. Make sure config.json for each model contains a "model_type" key, and that file permissions allow reading.

If you follow these steps and Docling still can't find your models, please share the value of DOCLING_SERVE_ARTIFACTS_PATH and a directory tree of your models folder so I can help debug further.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[egesnr]

@dosu thanks, now I will send upload this model folder to the drive should I zip it or tar ?

[dosubot[bot]]

Either zip or tar will work for uploading your model folder, but zip is usually the safest choice for cross-platform compatibility—Windows, macOS, and Linux all support extracting zip files natively. Tar (or tar.gz) is more common on Linux and preserves file permissions better, but Windows users may need extra tools to extract it.

If you don’t have special file permissions to preserve, zip is recommended. To create a zip archive from the command line:

zip -r docling_models.zip /path/to/your_models_dir

If you prefer tar (especially on Linux):

tar czvf docling_models.tar.gz /path/to/your_models_dir

Just make sure to extract the archive so the folder structure is preserved exactly as required for Docling.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[egesnr]

@dosu now after converting zip and send to the offline work env should i unzip to use or can i use like this import os
os.environ["DOCLING_SERVE_ARTIFACTS_PATH"] = "docling_models.zip"

[dosubot[bot]]

You need to unzip the archive first—Docling cannot use a zip file directly as the model artifacts path. Set DOCLING_SERVE_ARTIFACTS_PATH to the extracted directory containing your model subfolders, not to the zip file itself. The models must be accessible as regular files in the filesystem for Docling to load them properly.

So: unzip docling_models.zip, verify the folder structure is correct, then set the environment variable to the extracted directory path and run your code as usual. Details here.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}