Hugging Face · #176

docling-project/SmolDocling-256M-preview

docling-project/SmolDocling-256M-preview is a 256M image-text-to-text model on Hugging Face. Run locally: llama-cli -hf docling-project/SmolDocling-256M-preview or ollama run hf.co/docling-project/SmolDocling-256M-preview. 1,632 Hugging Face likes (#176 in this directory).

Model
https://huggingface.co/docling-project/SmolDocling-256M-preview
Author
docling-project (org)
Likes
1,632
Downloads (30 days)
30,040
Task
image-text-to-text
Library
transformers
License
cdla-permissive-2.0
Parameters
256M
Gated
No
Created
2025-02-12
Last modified
2025-09-17
Base models
HuggingFaceTB/SmolVLM-256M-Instruct
Snapshot
2026-10-29

Run it locally.

Shadowfetch Linux is a Debian testing desktop with KDE Plasma, local AI tools, and ShadowCode. ShadowCode's bundled llama.cpp runs GGUF models; you can also use llama.cpp or Ollama from the terminal.

llama.cpp

llama-cli -hf docling-project/SmolDocling-256M-preview

Ollama

ollama run hf.co/docling-project/SmolDocling-256M-preview

These commands work when this repo, or a GGUF quant of it, publishes GGUF weights. They do not invent a quant name. For safetensors-only repos, convert with llama.cpp or use a GGUF repo on the Hub.

ShadowCode prompt.

Paste this into ShadowCode after you open Settings › Local models.

In Settings › Local models, run docling-project/SmolDocling-256M-preview only if a GGUF file is available. If the Hub repo is safetensors-only, use a GGUF quant instead of guessing a filename. Ask before downloading.

Download Shadowfetch Linux — ISO, checksum and signature.

Back to Hugging Face models · Methodology