docling needs about 1.0 GB of memory at Q4_K_M (a 0.5 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 14 quants, from 0.4 GB (IQ4_NL) to 1.0 GB (F32).
Granite‑Docling 258M: a multimodal image‑to‑text conversion model
Granite‑Docling 258M is a multimodal image‑text‑to‑text model built on the Idefics3 architecture. It replaces the vision encoder with siglip2‑base‑patch16‑512 and uses a Granite 165M LLM. The model is licensed under Apache 2.0 and supports English, Japanese, Arabic and Chinese (experimental).
Good at
- Efficient document conversion
- Accurate equation recognition
- Flexible inference modes
Limits
- No vision support in other frameworks
- Limited to 8 k context length
- Requires at least 0.9 GB VRAM for IQ4_NL
Best for
- Batch PDF to Markdown conversion
- Extracting structured data from scanned forms
- Generating document summaries
Who it’s for. Developers building document processing pipelines who need a single model for image‑to‑text tasks. It integrates seamlessly with the Docling library and can be used with llama.cpp, transformers, vllm, or MLX.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ4_NL | 0.4 GB | 0.9 GB | Fits | Get file → | |
| Q4_0 | 0.4 GB | 0.9 GB | Fits | Get file → | |
| Q4_1 | 0.4 GB | 0.9 GB | Fits | Get file → | |
| Q5_0 | 0.4 GB | 0.9 GB | Fits | Get file → | |
| Q5_1 | 0.4 GB | 0.9 GB | Fits | Get file → | |
| Q8_0 | 0.4 GB | 1.0 GB | Fits | Get file → | |
| Q2_K | 0.5 GB | 1.0 GB | Fits | Get file → | |
| Q3_K_M | 0.5 GB | 1.0 GB | Fits | Get file → | |
| Q4_K_M | 0.5 GB | 1.0 GB | Fits | Get file → | |
| Q5_K_M | 0.5 GB | 1.0 GB | Fits | Get file → | |
| Q6_K | 0.5 GB | 1.0 GB | Fits | Get file → | |
| F16 | 0.5 GB | 1.1 GB | Fits | Get file → | |
| BF16 | 0.5 GB | 1.1 GB | Fits | Get file → | |
| F32 | 1.0 GB | 1.7 GB | Fits | Get file → |
For 8 GB VRAM, use IQ4_NL, Q4_0, or Q4_1 – file size 0.36 GB, memory 0.9 GB. If you have 12 GB VRAM, Q8_0 or Q6_K (0.42 GB, 1 GB memory) are suitable. For higher precision, choose F16 or BF16 (0.52 GB, 1.1 GB memory).
What GPU do you need for docling?
These figures add the file size, an fp16 KV cache at an 8K context window and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is F32 (1.0 GB file, about 1.7 GB in use).
Which docling quant should you choose?
Q4_K_M is the usual starting point for docling: it balances file size and quality well. Here it is a 0.5 GB file.
The Q8_0 file is 0.4 GB, about 0.9 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is IQ4_NL at 0.4 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about docling
How much VRAM does docling need?
About 1.0 GB at Q4_K_M with an 8K context window. That is the 0.5 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run docling on an 8 GB GPU?
Yes, with F32: the 1.0 GB file needs about 1.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of docling GGUF?
It depends on the quant: IQ4_NL is 0.4 GB, Q4_0 is 0.4 GB, Q4_1 is 0.4 GB, Q5_0 is 0.4 GB, Q5_1 is 0.4 GB, Q8_0 is 0.4 GB and 8 more.
Where can I download docling GGUF?
The files are on Hugging Face in the gguf-org/docling-gguf repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does docling use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run docling on 8 GB VRAM?
Yes, use IQ4_NL, Q4_0 or Q4_1; each requires about 0.9 GB VRAM at 8 k context.
Does docling support Japanese text?
Yes, Japanese support is experimental as noted in the model card.
What license applies to this model?
The model is released under the Apache 2.0 license.