Skip to content

Qianfan OCR VRAM requirements

How much VRAM Qianfan OCR needs
4-bit weights: 2.7 GBWith 8K context: 4.6 GBWith 32K context: 8.0 GB8-bit, 8K context: 6.6 GB
Parameters
4.7B
Longest context
32K
Released
Mar 2026
License
Apache 2.0
Strongest at
Reading documents

How much VRAM Qianfan OCR needs

Memory Qianfan OCR needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit8.8 GB10.8 GB14.2 GB
8-bit4.7 GB6.6 GB10.0 GB
6-bit3.6 GB5.6 GB8.9 GB
5-bit3.1 GB5.1 GB8.5 GB
4-bit2.7 GB4.6 GB8.0 GB
3-bit2.2 GB4.1 GB7.5 GB
2-bit1.7 GB3.6 GB7.0 GB
1.6-bit933 MB2.9 GB6.2 GB

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

GPUs and Macs that run Qianfan OCR

The best way Qianfan OCR runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40606-bit20K50 tok/s0.6 sPDF: Qianfan OCR on RTX 4060 report
RTX 50606-bit20K80 tok/s0.5 sPDF: Qianfan OCR on RTX 5060 report
RTX 30706-bit20K80 tok/s0.4 sPDF: Qianfan OCR on RTX 3070 report
RTX 3080 10 GB8-bit28K102 tok/s0.3 sPDF: Qianfan OCR on RTX 3080 10 GB report
RTX 3060 12 GB8-bit32K52 tok/s0.7 sPDF: Qianfan OCR on RTX 3060 12 GB report
Arc B5808-bit32K45 tok/s0.5 sPDF: Qianfan OCR on Arc B580 report
RTX 40708-bit32K71 tok/s0.3 sPDF: Qianfan OCR on RTX 4070 report
RTX 4070 Super8-bit32K71 tok/s0.3 sPDF: Qianfan OCR on RTX 4070 Super report
RTX 50708-bit32K92 tok/s0.3 sPDF: Qianfan OCR on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K42 tok/s0.4 sPDF: Qianfan OCR on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K63 tok/s0.4 sPDF: Qianfan OCR on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K54 tok/s0.4 sPDF: Qianfan OCR on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K72 tok/s0.2 sPDF: Qianfan OCR on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K92 tok/s0.2 sPDF: Qianfan OCR on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K100 tok/s0.2 sPDF: Qianfan OCR on RTX 4080 Super report
RTX 5070 Ti8-bit32K118 tok/s0.2 sPDF: Qianfan OCR on RTX 5070 Ti report
RTX 50808-bit32K126 tok/s0.2 sPDF: Qianfan OCR on RTX 5080 report
Radeon RX 7900 XTX8-bit32K102 tok/s0.4 sPDF: Qianfan OCR on Radeon RX 7900 XTX report
RTX 30908-bit32K123 tok/s0.3 sPDF: Qianfan OCR on RTX 3090 report
RTX 40908-bit32K131 tok/s0.1 sPDF: Qianfan OCR on RTX 4090 report
Radeon AI PRO R97008-bit32K72 tok/s0.3 sPDF: Qianfan OCR on Radeon AI PRO R9700 report
RTX 50908-bit32K210 tok/s0.1 sPDF: Qianfan OCR on RTX 5090 report
2× RTX 30908-bit32K123 tok/s0.1 sPDF: Qianfan OCR on 2× RTX 3090 report
2× RTX 40908-bit32K131 tok/s0.1 sPDF: Qianfan OCR on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K167 tok/s0.1 sPDF: Qianfan OCR on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K126 tok/s0.1 sPDF: Qianfan OCR on RTX 6000 Ada report
L40S8-bit32K115 tok/s0.2 sPDF: Qianfan OCR on L40S report
2× RTX 50908-bit32K210 tok/s0.1 sPDF: Qianfan OCR on 2× RTX 5090 report
4× RTX 30908-bit32K123 tok/s0.1 sPDF: Qianfan OCR on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K210 tok/s0.1 sPDF: Qianfan OCR on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit32K20 tok/s5.7 sPDF: Qianfan OCR on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K20 tok/s5.7 sPDF: Qianfan OCR on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K20 tok/s5.7 sPDF: Qianfan OCR on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K43 tok/s2.8 sPDF: Qianfan OCR on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K43 tok/s2.8 sPDF: Qianfan OCR on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K43 tok/s2.8 sPDF: Qianfan OCR on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K71 tok/s1.4 sPDF: Qianfan OCR on M4 Max 128 GB report
M5 Max 128 GB8-bit32K77 tok/s0.4 sPDF: Qianfan OCR on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K87 tok/s0.8 sPDF: Qianfan OCR on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K87 tok/s0.8 sPDF: Qianfan OCR on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K87 tok/s0.8 sPDF: Qianfan OCR on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K114 tok/s0.2 sPDF: Qianfan OCR on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K31 tok/s0.8 sPDF: Qianfan OCR on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K31 tok/s0.8 sPDF: Qianfan OCR on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K43 tok/s0.3 sPDF: Qianfan OCR on RTX Spark report
DGX Spark8-bit32K40 tok/s0.3 sPDF: Qianfan OCR on DGX Spark report
Dual DGX Spark8-bit32K40 tok/s0.2 sPDF: Qianfan OCR on Dual DGX Spark report
DGX Station GB3008-bit32K498 tok/s0.1 sPDF: Qianfan OCR on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K210 tok/s0.1 sPDF: Qianfan OCR on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K210 tok/s0.1 sPDF: Qianfan OCR on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K210 tok/s0.1 sPDF: Qianfan OCR on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K232 tok/s0.2 sPDF: Qianfan OCR on A100 80 GB SXM report
H100 SXM8-bit32K328 tok/s0.1 sPDF: Qianfan OCR on H100 SXM report
4× H200 NVL8-bit32K407 tok/s0.1 sPDF: Qianfan OCR on 4× H200 NVL report
8× H200 NVL8-bit32K407 tok/s0.1 sPDF: Qianfan OCR 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 Reza2kn/Qianfan-OCR-GGUF
vLLM
vllm serve baidu/Qianfan-OCR

Questions

How much VRAM does Qianfan OCR need?

Qianfan OCR needs 2.7 GB for its weights at 4-bit, 4.6 GB with an 8K context and 8.0 GB with 32K. At 8-bit, 6.6 GB.

What is the minimum VRAM for Qianfan OCR?

About 4.1 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 Qianfan OCR?

4.7B parameters: 8.8 GB of weights at 16-bit, 2.7 GB at 4-bit.

Can Qianfan OCR run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 131 tokens a second.

More models