# Unlimited-OCR GPU requirements

Unlimited-OCR has 3.3B parameters. To serve it, plan for about 8.0 GB of GPU memory in 16-bit, 4.0 GB in 8-bit or 2.0 GB in 4-bit, including about 20% for the KV cache and activations.

It is a mixture of experts: 64 experts, 6 of them active for each token. All of them stay in GPU memory, so plan memory for all 3.3B parameters.

[Deploy it on Arrakis](/console/#/models?model=baidu/Unlimited-OCR).

## What it takes to serve it

| Precision | Memory and the smallest setup |
|---|---|
| 16-bit | 8.0 GB: one [RTX 4090](/gpus/rtx-4090.html.md) or [RTX 3090](/gpus/rtx-3090.html.md), RTX 3090 from $0.19 per GPU-hour |
| 8-bit | 4.0 GB: one [RTX 4090](/gpus/rtx-4090.html.md) or [RTX 3090](/gpus/rtx-3090.html.md), RTX 3090 from $0.19 per GPU-hour |
| 4-bit | 2.0 GB: one [RTX 4090](/gpus/rtx-4090.html.md) or [RTX 3090](/gpus/rtx-3090.html.md), RTX 3090 from $0.19 per GPU-hour |

## What it takes to fine-tune it

| Method | Memory and the smallest setup |
|---|---|
| QLoRA | 2.3 GB: one [RTX 4090](/gpus/rtx-4090.html.md) or [RTX 3090](/gpus/rtx-3090.html.md), RTX 3090 from $0.19 per GPU-hour |
| LoRA | 8.0 GB: one [RTX 4090](/gpus/rtx-4090.html.md) or [RTX 3090](/gpus/rtx-3090.html.md), RTX 3090 from $0.19 per GPU-hour |
| Full fine-tuning | 53 GB: one [H100](/gpus/h100.html.md) or [A100](/gpus/a100.html.md), H100 from $1.65 per GPU-hour |

## About the model

| | |
|---|---|
| Parameters | `3.3B` |
| Experts | `64` |
| Experts for each token | `6` |
| Context window (tokens) | `32,768` |
| Published in | `BF16` |
| License | `mit` |
| Released | `2026-06-19` |
| Weights | [baidu/Unlimited-OCR](https://huggingface.co/baidu/Unlimited-OCR) |

## Serve it yourself

```
vllm serve baidu/Unlimited-OCR --tensor-parallel-size 1
```

## Questions

**How much GPU memory does Unlimited-OCR need?** To serve it, about 8 GB in 16-bit, 4 GB in 8-bit, 2 GB in 4-bit, including about 20% for the KV cache and activations. To fine-tune it, about 2 GB with QLoRA, 8 GB with LoRA and 53 GB for full fine-tuning with Adam, before activations.

**Can I run Unlimited-OCR on one GPU?** Yes. In 16-bit it needs about 8.0 GB, which fits on one RTX 4090 or RTX 3090.

**What does it cost to run Unlimited-OCR?** In 16-bit it needs one RTX 4090 or RTX 3090; in 4-bit, one RTX 4090 or RTX 3090. Each setup on this page shows the lowest price per GPU-hour on Arrakis right now, and you pay by the hour from a prepaid balance.

**How do I deploy Unlimited-OCR on Arrakis?** Open Models in the Arrakis console, paste baidu/Unlimited-OCR, choose vLLM, SGLang or llama.cpp, and rent a GPU it fits on. You get an OpenAI-compatible endpoint with its own API key, and root SSH to the machine.


More models: [/models](/models.html.md).