Replies: 5 comments 8 replies
-
|
Hi @jncoppeta! I'm Dosu and I’m helping the docling team. Your issue is likely due to a combination of the environment variable used and the directory structure expected by Docling for offline model loading. First, Docling expects the environment variable to be or whatever the absolute path is to your models directory, and use this variable in your environment (including inside containers if applicable) citation. Second, the directory structure under your artifacts path must match what Docling expects. For layout models, Docling looks for a subfolder like For EasyOCR, Docling expects the model file at: If you see a "downloads disabled" error, it means the file is missing and Docling is not allowed to fetch it, so you must ensure it is present at that exact path citation. If you moved files from Finally, make sure you have downloaded all required models for your pipelines (including OCR, layout, and any others you use) using: and that If you follow these steps and ensure the directory structure and environment variable are correct, Docling should work offline without missing file errors. To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other |
Beta Was this translation helpful? Give feedback.
-
|
@dosu DOCLING_SERVE_ARTIFACTS_PATH is also set tup This is the result in the docling folder after the download: |
Beta Was this translation helpful? Give feedback.
-
|
Your top-level directory structure looks correct, but Docling requires specific files inside these subfolders for offline mode to work. For example, under ds4sd--docling-models you need model.safetensors, preprocessor_config.json, and config.json, and for EasyOCR you need ds4sd--docling-models/EasyOcr/craft_mlt_25k.pth and related files. If any of these are missing, you'll get missing file errors and Docling won't attempt to download them if downloads are disabled (source). To troubleshoot further, please run a recursive directory listing (e.g., If you find any missing files, re-run To reply, just mention @dosu. How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other |
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
-
|
I am encountering the same issue. In the folder I tried to fetch the models by running I also checked the source Hugging Face repository and confirmed that model.safetensors is not present in the file listing: https://huggingface.co/ds4sd/docling-models/tree/main. |
Beta Was this translation helpful? Give feedback.
-
|
In past I've got this error couple of times when models where downloaded with older version of docling and model cache was accessed with newer version. docling/docs/usage/advanced_options.md Lines 37 to 42 in df60673 @jncoppeta which version of Docling you are using? @FernandoD95 you can specify particular models and specific folder for models like this: then use the same folder (/modelcache in this example) as artifacts_path variable for the conversion options. There are several models that get downloaded by default and they are in different Hugging Face repos, some have |
Beta Was this translation helpful? Give feedback.
-
|
I see what you mean. The error message might be misleading, the Let me share an example of using utility, you can see which models are pulled: As you can see, the download utility points to the |
Beta Was this translation helpful? Give feedback.
-
|
My problem is that even when I do this correctly, the error still occurs. My folder structure is exactly like the one in your example. When I configure my Python script to point to that parent folder, I get the Here is the exact code I'm running and the error it produces: # The path is set to the parent directory containing all model folders
artifacts_path = "./modelscache"
pipeline_options = PdfPipelineOptions(artifacts_path=artifacts_path)
doc_converter = DocumentConverter(
format_options={
InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)
}
)
# ... when I run the conversion ...
# It fails with this error:
# FileNotFoundError: Missing safe tensors file: './modelscache/model.safetensors'It seems the |
Beta Was this translation helpful? Give feedback.
-
|
@FernandoD95 try fully resolved path. |
Beta Was this translation helpful? Give feedback.
-
|
You've pinpointed the issue! The problem was indeed the version. I was running on My apologies, I should have checked for a version update sooner. Thank you so much! |
Beta Was this translation helpful? Give feedback.
-
|
No worries at all! |
Beta Was this translation helpful? Give feedback.
-
|
@jncoppeta the default value of "artifacts_path" is different on different platforms and I'm not 100% sure if relative path work correctly.
|
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
I created a RAG ingestion pipeline around Docling and am now trying to convert it to work offline. I followed the guide outlines at https://docling-project.github.io/docling/usage/advanced_options/ and have not had success. Looks like the default paths are not configured correctly when the DOCLING_ARTIFACTS_PATH environment varible is set:
For context I downloaded the models, moved them from
~/.cacheto/app/.cache, and setDOCLING_ARTIFACTS_PATH= /app/.cache/docling/models. If I change the path to/app/.cache/docling/models/ds4sd--docling-modelsI get a different error:FileNotFoundError: Missing /app/.cache/docling/models/ds4sd--docling-models/EasyOcr/craft_mlt_25k.pth and downloads disabledAnyone encountered this before?
Beta Was this translation helpful? Give feedback.
All reactions