Skip to content

GOT OCR2 0 VRAM requirements

How much VRAM GOT OCR2 0 needs
4-bit weights: 414 MBWith 8K context: 1.9 GBWith 32K context: 4.2 GB8-bit, 8K context: 2.2 GB
Parameters
716M
Longest context
32K
Released
Sep 2024
License
Apache 2.0
Strongest at
Text and images

How much VRAM GOT OCR2 0 needs

Memory GOT OCR2 0 needs at each precision and context length
PrecisionWeightsWith 8KWith 32K
16-bit1.3 GB2.9 GB5.1 GB
8-bit726 MB2.2 GB4.5 GB
6-bit560 MB2.1 GB4.3 GB
5-bit486 MB2.0 GB4.2 GB
4-bit414 MB1.9 GB4.2 GB
3-bit334 MB1.8 GB4.1 GB
2-bit256 MB1.8 GB4.0 GB
1.6-bit141 MB1.7 GB3.9 GB

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

GPUs and Macs that run GOT OCR2 0

The best way GOT OCR2 0 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit32K212 tok/s0.1 sPDF: GOT OCR2 0 on RTX 4060 report
RTX 50608-bit32K318 tok/s0.1 sPDF: GOT OCR2 0 on RTX 5060 report
RTX 30708-bit32K318 tok/s0.1 sPDF: GOT OCR2 0 on RTX 3070 report
RTX 3080 10 GB8-bit32K465 tok/s0.1 sPDF: GOT OCR2 0 on RTX 3080 10 GB report
RTX 3060 12 GB8-bit32K267 tok/s0.1 sPDF: GOT OCR2 0 on RTX 3060 12 GB report
Arc B5808-bit32K209 tok/s0.1 sPDF: GOT OCR2 0 on Arc B580 report
RTX 40708-bit32K348 tok/s0.1 sPDF: GOT OCR2 0 on RTX 4070 report
RTX 4070 Super8-bit32K348 tok/s0.1 sPDF: GOT OCR2 0 on RTX 4070 Super report
RTX 50708-bit32K428 tok/s0.1 sPDF: GOT OCR2 0 on RTX 5070 report
RTX 4060 Ti 16 GB8-bit32K223 tok/s0.1 sPDF: GOT OCR2 0 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit32K318 tok/s0.1 sPDF: GOT OCR2 0 on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit32K241 tok/s0.1 sPDF: GOT OCR2 0 on Arc A770 16 GB report
Radeon RX 9070 XT8-bit32K314 tok/s0.1 sPDF: GOT OCR2 0 on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit32K428 tok/s0.1 sPDF: GOT OCR2 0 on RTX 4070 Ti Super report
RTX 4080 Super8-bit32K455 tok/s0.1 sPDF: GOT OCR2 0 on RTX 4080 Super report
RTX 5070 Ti8-bit32K517 tok/s0.1 sPDF: GOT OCR2 0 on RTX 5070 Ti report
RTX 50808-bit32K540 tok/s0.1 sPDF: GOT OCR2 0 on RTX 5080 report
Radeon RX 7900 XTX8-bit32K397 tok/s0.1 sPDF: GOT OCR2 0 on Radeon RX 7900 XTX report
RTX 30908-bit32K532 tok/s0.1 sPDF: GOT OCR2 0 on RTX 3090 report
RTX 40908-bit32K556 tok/s0.1 sPDF: GOT OCR2 0 on RTX 4090 report
Radeon AI PRO R97008-bit32K314 tok/s0.1 sPDF: GOT OCR2 0 on Radeon AI PRO R9700 report
RTX 50908-bit32K754 tok/s0.1 sPDF: GOT OCR2 0 on RTX 5090 report
2× RTX 30908-bit32K532 tok/s0.1 sPDF: GOT OCR2 0 on 2× RTX 3090 report
2× RTX 40908-bit32K556 tok/s0.1 sPDF: GOT OCR2 0 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit32K654 tok/s0.1 sPDF: GOT OCR2 0 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit32K540 tok/s0.1 sPDF: GOT OCR2 0 on RTX 6000 Ada report
L40S8-bit32K506 tok/s0.1 sPDF: GOT OCR2 0 on L40S report
2× RTX 50908-bit32K754 tok/s0.1 sPDF: GOT OCR2 0 on 2× RTX 5090 report
4× RTX 30908-bit32K532 tok/s0.1 sPDF: GOT OCR2 0 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit32K754 tok/s0.1 sPDF: GOT OCR2 0 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit32K104 tok/s0.8 sPDF: GOT OCR2 0 on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit32K104 tok/s0.8 sPDF: GOT OCR2 0 on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit32K104 tok/s0.8 sPDF: GOT OCR2 0 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit32K188 tok/s0.4 sPDF: GOT OCR2 0 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit32K188 tok/s0.4 sPDF: GOT OCR2 0 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit32K188 tok/s0.4 sPDF: GOT OCR2 0 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit32K265 tok/s0.2 sPDF: GOT OCR2 0 on M4 Max 128 GB report
M5 Max 128 GB8-bit32K280 tok/s0.1 sPDF: GOT OCR2 0 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit32K299 tok/s0.1 sPDF: GOT OCR2 0 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit32K299 tok/s0.1 sPDF: GOT OCR2 0 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit32K299 tok/s0.1 sPDF: GOT OCR2 0 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit32K345 tok/s0.1 sPDF: GOT OCR2 0 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit32K163 tok/s0.1 sPDF: GOT OCR2 0 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit32K163 tok/s0.1 sPDF: GOT OCR2 0 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit32K230 tok/s0.1 sPDF: GOT OCR2 0 on RTX Spark report
DGX Spark8-bit32K213 tok/s0.1 sPDF: GOT OCR2 0 on DGX Spark report
Dual DGX Spark8-bit32K213 tok/s0.1 sPDF: GOT OCR2 0 on Dual DGX Spark report
DGX Station GB3008-bit32K1,145 tok/s0.1 sPDF: GOT OCR2 0 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit32K754 tok/s0.1 sPDF: GOT OCR2 0 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit32K754 tok/s0.1 sPDF: GOT OCR2 0 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit32K754 tok/s0.1 sPDF: GOT OCR2 0 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit32K798 tok/s0.1 sPDF: GOT OCR2 0 on A100 80 GB SXM report
H100 SXM8-bit32K957 tok/s0.1 sPDF: GOT OCR2 0 on H100 SXM report
4× H200 NVL8-bit32K1,057 tok/s0.1 sPDF: GOT OCR2 0 on 4× H200 NVL report
8× H200 NVL8-bit32K1,057 tok/s0.1 sPDF: GOT OCR2 0 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 cstr/got-ocr2-crispembed-GGUF
vLLM
vllm serve stepfun-ai/GOT-OCR2_0
MLX (Mac)
mlx_lm.chat --model mlx-community/GOT-OCR2_0-4bit

Questions

How much VRAM does GOT OCR2 0 need?

GOT OCR2 0 needs 414 MB for its weights at 4-bit, 1.9 GB with an 8K context and 4.2 GB with 32K. At 8-bit, 2.2 GB.

What is the minimum VRAM for GOT OCR2 0?

About 1.8 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 GOT OCR2 0?

716M parameters: 1.3 GB of weights at 16-bit, 414 MB at 4-bit.

Can GOT OCR2 0 run on an RTX 4090?

Yes. At 8-bit it holds 32K of context and writes about 556 tokens a second.

More models