datalab-toQwen3_5ForConditionalGeneration

Surya Ocr 2.5 — Hardware Requirements & GPU Compatibility

Vision

Surya 2.5 is an OCR toolkit from Datalab for document intelligence. A single small vision-language model with about 689 million parameters handles full-page text recognition, layout analysis with reading order, math, and tables output as HTML, with support for many languages. A separate lightweight model detects text lines. It follows Surya 2 and uses a Qwen3.5-style architecture. Context length is 262,144 tokens. The code is Apache 2.0, while the model weights use a modified OpenRAIL-M license that is free for research, personal use, and startups under $5M in funding or revenue, with a separate commercial license for broader use. The model is served through vLLM on an NVIDIA GPU or llama.cpp on CPU or Apple Silicon, and at this size it runs on a consumer GPU or a laptop.

689M params · dense262K contextopenrailOct 202635.0K downloads49 likes62.8K quant downloads

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

Runs great on a 24 GB GPU

Runs from 8 GB (Q4_K_M · 256K)Best on 24 GB (Q4_K_M · ~862 tok/s · 256K)Macs from 8 GB (Q4_K_M · ~94 tok/s)

openrail

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Surya Ocr 2.5

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Ways to runOther ways

each opens step-by-step

Run in cloud

Runs on RTX 3060 12GB (11 GB headroom) · Q4_K_M

Generation speed: ~308 tok/s; generation speed
Cost per 1M output tokens: $0.055; per 1M output tokens

Deployon Vast.aiBest valuefrom $0.061/hr

Compare GPUs →

Run Surya Ocr 2.5

0.8 GB VRAM at 16K context · fits the RTX 4090 (23.2 GB free)

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/datalab-to/surya-ocr-2.5-gguf
datalab-to/surya-ocr-2.5-gguf does not list Q4_K_M; Ollama picks a quant itself (Q4_K_M if present). Check the repo's Files tab.
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/datalab-to/surya-ocr-2.5-gguf, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/datalab-to/surya-ocr-2.5-gguf
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf datalab-to/surya-ocr-2.5-gguf -c 16384 -ngl 99 --port 8080
datalab-to/surya-ocr-2.5-gguf does not list Q4_K_M; llama.cpp falls back to Q4_K_M or the first file.

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/datalab-to/surya-ocr-2.5-gguf
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf datalab-to/surya-ocr-2.5-gguf -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve datalab-to/surya-ocr-2.5 --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~2 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model kltng/surya-ocr-2.5-mlx-8bit
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model kltng/surya-ocr-2.5-mlx-8bit

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$datalab-to/surya-ocr-2.5-gguf
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
RTX 3060 12GBCheapestCheapest11 GB free	~308 tok/s	$0.055	Deploy on Vast.aifrom $0.061/hr (opens in new tab)	11 GB free · ~308 tok/s · $0.055 / 1M tokens
RTX 3060 Ti (8 GB)7 GB free	~383 tok/s	$0.050	Deploy on Vast.aifrom $0.069/hr (opens in new tab)	8 GB · 7 GB free · ~383 tok/s · $0.050 / 1M tokens
RTX 4060 (8 GB)7 GB free	~233 tok/s	$0.099	Deploy on Vast.aifrom $0.083/hr (opens in new tab)	8 GB · 7 GB free · ~233 tok/s · $0.099 / 1M tokens
RTX A4000 (16 GB)15 GB free	~383 tok/s	$0.064	Deploy on Vast.aifrom $0.088/hr (opens in new tab)	16 GB · 15 GB free · ~383 tok/s · $0.064 / 1M tokens
RTX A6000 (48 GB)47 GB free	~657 tok/s	$0.14	Deploy on RunPodfrom $0.33/hr (opens in new tab)	48 GB · 47 GB free · ~657 tok/s · $0.14 / 1M tokens
RTX PRO 4500 (32 GB)31 GB free	~766 tok/s	$0.12	Deploy on RunPodfrom $0.34/hr (opens in new tab)	32 GB · 31 GB free · ~766 tok/s · $0.12 / 1M tokens
RTX 4090 (24 GB)23 GB free	~862 tok/s	$0.11	Deploy on RunPodfrom $0.34/hr (opens in new tab)	24 GB · 23 GB free · ~862 tok/s · $0.11 / 1M tokens
A40 (48 GB)47 GB free	~595 tok/s	$0.16	Deploy on RunPodfrom $0.35/hr (opens in new tab)	48 GB · 47 GB free · ~595 tok/s · $0.16 / 1M tokens

Compare GPUs →

How Much VRAM Does Surya Ocr 2.5 Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
0.6 GB	0.8 GB	1.4 GB	~0.3 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Fits: 256K context · ~1024 tok/s
0.7 GB	0.8 GB	1.4 GB	~0.3 GB	3-bit medium — 3-bit medium quantization	Fits: 256K context · ~950 tok/s
0.8 GB	0.9 GB	1.5 GB	~0.4 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Fits: 256K context · ~862 tok/s
0.8 GB	1 GB	1.6 GB	~0.5 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Fits: 256K context · ~780 tok/s
0.9 GB	1.1 GB	1.7 GB	~0.6 GB	6-bit, very good — 6-bit quantization, very good quality	Fits: 256K context · ~712 tok/s
1.0 GB	1.2 GB	1.8 GB	0.6 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Fits: 256K context · ~630 tok/s
1.7 GB	1.9 GB	2.5 GB	~1.4 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Fits: 256K context · ~379 tok/s

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

0.8GBat 16K context

Weights: 0.4 GB
KV cache: 0.1 GB
Overhead: 0.3 GB

Fits RTX 4090 (reference) · 23.2 GB free · ~862 tok/s

≥50 tok/s needs ≥58 GB/s · RTX 5090 ~1533, RTX 3090 Ti ~862

Which Hardware Can Run Surya Ocr 2.5?

Q4_K_M · 0.8 GB

Surya Ocr 2.5 (Q4_K_M) requires 0.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 1+ GB is recommended. Using the full 262K context window can add up to 1.6 GB, bringing total usage to 2.3 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run Surya Ocr 2.5, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8 GB · RTX 4060Q4_K_M · 0.8 GB · ~233 tok/s · 256K0.8~233256K8 GB tier →

12 GB · RTX 3060 12GBQ4_K_M · 0.8 GB · ~308 tok/s · 256K0.8~308256K12 GB tier →

16 GB · RTX 5060 Ti 16GBQ4_K_M · 0.8 GB · ~383 tok/s · 256K0.8~383256K16 GB tier →

24 GB · RTX 4090referenceQ4_K_M · 0.8 GB · ~862 tok/s · 256K0.8~862256K24 GB tier →

Mac 32 GB · Mac Mini M4Q4_K_M · 0.8 GB · ~111 tok/s · 256K0.8~111256KMacs →

Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 0.8 GB · ~503 tok/s · 256K0.8~503256KMacs →

Specifications

Publisher: datalab-to
Parameters: 689M
Architecture: Qwen3_5ForConditionalGeneration
Context Length: 262,144 tokens
Vocabulary Size: 88,526
Release Date: 2026-10-25
License: openrail

Resources

Hugging Face model card (opens in a new tab)

Frequently Asked Questions

How much VRAM does Surya Ocr 2.5 need?

Surya Ocr 2.5 requires 0.8 GB of VRAM at Q4_K_M, or 1.7 GB at BF16. Full 262K context adds up to 1.6 GB (2.3 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 689M × 4.8 bits ÷ 8 = 0.4 GB

KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 1.9 GB (at full 262K context)

VRAM usage by quantization

Q4_K_M

0.8 GB

Q4_K_M + full context

2.3 GB

Learn more about VRAM estimation →

What's the best quantization for Surya Ocr 2.5?

For Surya Ocr 2.5, Q4_K_M (0.8 GB) offers the best balance of quality and VRAM usage. Q5_K_M (0.8 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 0.6 GB.

VRAM requirement by quantization

Q2_K

0.6 GB~75%

Q4_K_M ★

0.8 GB~89%

Q5_K_M

0.8 GB~92%

Q6_K

0.9 GB~95%

Q8_0

1.0 GB~99%

BF16

1.7 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Surya Ocr 2.5 on a Mac?

Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run Surya Ocr 2.5. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 0.8 GB of usable unified memory (RAM minus macOS overhead).

Can I run Surya Ocr 2.5 locally?

Yes — Surya Ocr 2.5 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Surya Ocr 2.5?

At Q4_K_M, Surya Ocr 2.5 can reach ~6316 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~862 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 0.76 × 0.65 = ~6842 tok/s

Estimated speed at Q4_K_M (0.8 GB)

NVIDIA B200

~6842 tok/s

NVIDIA GeForce RTX 4090

~862 tok/s

NVIDIA B300

~6842 tok/s

AMD Instinct MI350X

~6316 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Surya Ocr 2.5?

At Q4_K_M, the download is about 0.41 GB. The full-precision BF16 version is 1.38 GB. The smallest option (Q2_K) is 0.29 GB.

Which GPUs can run Surya Ocr 2.5?

52 consumer GPUs can run Surya Ocr 2.5 at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.

Which devices can run Surya Ocr 2.5?

163 devices with unified memory can run Surya Ocr 2.5 at Q4_K_M (0.8 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.