Skip to content

MinerU2.5 2509 1.2B VRAM requirements

How much VRAM MinerU2.5 2509 1.2B needs
4-bit weights: 668 MBWith 8K context: 1.5 GBWith 16K context: 1.6 GB8-bit, 8K context: 2.0 GB
Parameters
1.2B
Longest context
16K
Released
Sep 2025
License
agpl-3.0
Strongest at
Text and images

Its licence (agpl-3.0) is not a standard open-source one: check the terms before commercial use.

How much VRAM MinerU2.5 2509 1.2B needs

Memory MinerU2.5 2509 1.2B needs at each precision and context length
PrecisionWeightsWith 8KWith 16K
16-bit2.2 GB3.0 GB3.1 GB
8-bit1.1 GB2.0 GB2.1 GB
6-bit904 MB1.7 GB1.8 GB
5-bit784 MB1.6 GB1.7 GB
4-bit668 MB1.5 GB1.6 GB
3-bit539 MB1.4 GB1.5 GB
2-bit413 MB1.3 GB1.4 GB
1.6-bit227 MB1.1 GB1.2 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run MinerU2.5 2509 1.2B

The best way MinerU2.5 2509 1.2B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit16K139 tok/s0.2 sPDF: MinerU2.5 2509 1.2B on RTX 4060 report
RTX 50608-bit16K216 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 5060 report
RTX 30708-bit16K216 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 3070 report
RTX 3080 10 GB8-bit16K330 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit16K179 tok/s0.2 sPDF: MinerU2.5 2509 1.2B on RTX 3060 12 GB report
Arc B5808-bit16K146 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on Arc B580 report
RTX 40708-bit16K238 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 4070 report
RTX 4070 Super8-bit16K238 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 4070 Super report
RTX 50708-bit16K300 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit16K147 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit16K216 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit16K172 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit16K227 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit16K300 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 4070 Ti Super report
RTX 4080 Super8-bit16K322 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 4080 Super report
RTX 5070 Ti8-bit16K373 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 5070 Ti report
RTX 50808-bit16K392 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 5080 report
Radeon RX 7900 XTX8-bit16K300 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on Radeon RX 7900 XTX report
RTX 30908-bit16K385 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 3090 report
RTX 40908-bit16K406 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 4090 report
Radeon AI PRO R97008-bit16K227 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on Radeon AI PRO R9700 report
RTX 50908-bit16K589 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 5090 report
2× RTX 30908-bit16K385 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on 2× RTX 3090 report
2× RTX 40908-bit16K406 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit16K494 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit16K392 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX 6000 Ada report
L40S8-bit16K364 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on L40S report
2× RTX 50908-bit16K589 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on 2× RTX 5090 report
4× RTX 30908-bit16K385 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit16K589 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit16K69 tok/s1.5 sPDF: MinerU2.5 2509 1.2B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit16K69 tok/s1.5 sPDF: MinerU2.5 2509 1.2B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit16K69 tok/s1.5 sPDF: MinerU2.5 2509 1.2B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit16K135 tok/s0.7 sPDF: MinerU2.5 2509 1.2B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit16K135 tok/s0.7 sPDF: MinerU2.5 2509 1.2B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit16K135 tok/s0.7 sPDF: MinerU2.5 2509 1.2B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit16K203 tok/s0.4 sPDF: MinerU2.5 2509 1.2B on M4 Max 128 GB report
M5 Max 128 GB8-bit16K218 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit16K237 tok/s0.2 sPDF: MinerU2.5 2509 1.2B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit16K237 tok/s0.2 sPDF: MinerU2.5 2509 1.2B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit16K237 tok/s0.2 sPDF: MinerU2.5 2509 1.2B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit16K286 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit16K109 tok/s0.2 sPDF: MinerU2.5 2509 1.2B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit16K109 tok/s0.2 sPDF: MinerU2.5 2509 1.2B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit16K152 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on RTX Spark report
DGX Spark8-bit16K140 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on DGX Spark report
Dual DGX Spark8-bit16K140 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on Dual DGX Spark report
DGX Station GB3008-bit16K1,034 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit16K589 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit16K589 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit16K589 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit16K633 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on A100 80 GB SXM report
H100 SXM8-bit16K804 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on H100 SXM report
4× H200 NVL8-bit16K921 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on 4× H200 NVL report
8× H200 NVL8-bit16K921 tok/s0.1 sPDF: MinerU2.5 2509 1.2B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf mradermacher/MinerU2.5-2509-1.2B-GGUF
vLLM
vllm serve opendatalab/MinerU2.5-2509-1.2B

Questions

How much VRAM does MinerU2.5 2509 1.2B need?

MinerU2.5 2509 1.2B needs 668 MB for its weights at 4-bit, 1.5 GB with an 8K context and 1.6 GB with 16K. At 8-bit, 2.0 GB.

What is the minimum VRAM for MinerU2.5 2509 1.2B?

About 1.4 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is MinerU2.5 2509 1.2B?

1.2B parameters: 2.2 GB of weights at 16-bit, 668 MB at 4-bit.

Can MinerU2.5 2509 1.2B run on an RTX 4090?

Yes. At 8-bit it holds 16K of context and writes about 406 tokens a second.

More models