Skip to content

Qwen2.5 Math 7B Instruct VRAM requirements

How much VRAM Qwen2.5 Math 7B Instruct needs
4-bit weights: 4.3 GBWith 4K context: 5.4 GB8-bit, 4K context: 8.6 GB
Parameters
7.6B
Longest context
4K
Released
Sep 2024
License
Apache 2.0
Strongest at
Not rated

How much VRAM Qwen2.5 Math 7B Instruct needs

Memory Qwen2.5 Math 7B Instruct needs at each precision and context length
PrecisionWeightsWith 4K
16-bit14.2 GB15.3 GB
8-bit7.5 GB8.6 GB
6-bit5.8 GB6.9 GB
5-bit5.0 GB6.1 GB
4-bit4.3 GB5.4 GB
3-bit3.5 GB4.5 GB
2-bit2.7 GB3.7 GB
1.6-bit1.5 GB2.5 GB

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

GPUs and Macs that run Qwen2.5 Math 7B Instruct

The best way Qwen2.5 Math 7B Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40606-bit4K32 tok/s1.0 sPDF: Qwen2.5 Math 7B Instruct on RTX 4060 report
RTX 50606-bit4K52 tok/s0.8 sPDF: Qwen2.5 Math 7B Instruct on RTX 5060 report
RTX 30706-bit4K52 tok/s0.7 sPDF: Qwen2.5 Math 7B Instruct on RTX 3070 report
RTX 3080 10 GB8-bit4K67 tok/s0.5 sPDF: Qwen2.5 Math 7B Instruct on RTX 3080 10 GB report
RTX 3060 12 GB8-bit4K33 tok/s1.1 sPDF: Qwen2.5 Math 7B Instruct on RTX 3060 12 GB report
Arc B5808-bit4K29 tok/s0.8 sPDF: Qwen2.5 Math 7B Instruct on Arc B580 report
RTX 40708-bit4K45 tok/s0.5 sPDF: Qwen2.5 Math 7B Instruct on RTX 4070 report
RTX 4070 Super8-bit4K45 tok/s0.4 sPDF: Qwen2.5 Math 7B Instruct on RTX 4070 Super report
RTX 50708-bit4K59 tok/s0.5 sPDF: Qwen2.5 Math 7B Instruct on RTX 5070 report
RTX 4060 Ti 16 GB8-bit4K26 tok/s0.7 sPDF: Qwen2.5 Math 7B Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit4K40 tok/s0.6 sPDF: Qwen2.5 Math 7B Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit4K35 tok/s0.7 sPDF: Qwen2.5 Math 7B Instruct on Arc A770 16 GB report
Radeon RX 9070 XT8-bit4K48 tok/s0.4 sPDF: Qwen2.5 Math 7B Instruct on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit4K59 tok/s0.3 sPDF: Qwen2.5 Math 7B Instruct on RTX 4070 Ti Super report
RTX 4080 Super8-bit4K65 tok/s0.3 sPDF: Qwen2.5 Math 7B Instruct on RTX 4080 Super report
RTX 5070 Ti8-bit4K78 tok/s0.3 sPDF: Qwen2.5 Math 7B Instruct on RTX 5070 Ti report
RTX 50808-bit4K83 tok/s0.3 sPDF: Qwen2.5 Math 7B Instruct on RTX 5080 report
Radeon RX 7900 XTX8-bit4K69 tok/s0.6 sPDF: Qwen2.5 Math 7B Instruct on Radeon RX 7900 XTX report
RTX 30908-bit4K81 tok/s0.4 sPDF: Qwen2.5 Math 7B Instruct on RTX 3090 report
RTX 40908-bit4K87 tok/s0.2 sPDF: Qwen2.5 Math 7B Instruct on RTX 4090 report
Radeon AI PRO R97008-bit4K48 tok/s0.4 sPDF: Qwen2.5 Math 7B Instruct on Radeon AI PRO R9700 report
RTX 50908-bit4K146 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on RTX 5090 report
2× RTX 30908-bit4K81 tok/s0.2 sPDF: Qwen2.5 Math 7B Instruct on 2× RTX 3090 report
2× RTX 40908-bit4K87 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit4K113 tok/s0.2 sPDF: Qwen2.5 Math 7B Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit4K83 tok/s0.2 sPDF: Qwen2.5 Math 7B Instruct on RTX 6000 Ada report
L40S8-bit4K75 tok/s0.3 sPDF: Qwen2.5 Math 7B Instruct on L40S report
2× RTX 50908-bit4K146 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on 2× RTX 5090 report
4× RTX 30908-bit4K81 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit4K146 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit4K13 tok/s9.2 sPDF: Qwen2.5 Math 7B Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit4K13 tok/s9.2 sPDF: Qwen2.5 Math 7B Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit4K13 tok/s9.2 sPDF: Qwen2.5 Math 7B Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit4K28 tok/s4.6 sPDF: Qwen2.5 Math 7B Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit4K28 tok/s4.6 sPDF: Qwen2.5 Math 7B Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit4K28 tok/s4.6 sPDF: Qwen2.5 Math 7B Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit4K49 tok/s2.3 sPDF: Qwen2.5 Math 7B Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit4K54 tok/s0.7 sPDF: Qwen2.5 Math 7B Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit4K61 tok/s1.3 sPDF: Qwen2.5 Math 7B Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit4K61 tok/s1.3 sPDF: Qwen2.5 Math 7B Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit4K61 tok/s1.3 sPDF: Qwen2.5 Math 7B Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit4K84 tok/s0.4 sPDF: Qwen2.5 Math 7B Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit4K20 tok/s1.3 sPDF: Qwen2.5 Math 7B Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit4K20 tok/s1.3 sPDF: Qwen2.5 Math 7B Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit4K27 tok/s0.5 sPDF: Qwen2.5 Math 7B Instruct on RTX Spark report
DGX Spark8-bit4K25 tok/s0.5 sPDF: Qwen2.5 Math 7B Instruct on DGX Spark report
Dual DGX Spark8-bit4K25 tok/s0.3 sPDF: Qwen2.5 Math 7B Instruct on Dual DGX Spark report
DGX Station GB3008-bit4K425 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit4K146 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit4K146 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit4K146 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit4K164 tok/s0.3 sPDF: Qwen2.5 Math 7B Instruct on A100 80 GB SXM report
H100 SXM8-bit4K247 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on H100 SXM report
4× H200 NVL8-bit4K325 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct on 4× H200 NVL report
8× H200 NVL8-bit4K325 tok/s0.1 sPDF: Qwen2.5 Math 7B Instruct 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 bartowski/Qwen2.5-Math-7B-Instruct-GGUF
vLLM
vllm serve Qwen/Qwen2.5-Math-7B-Instruct
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen2.5-Math-7B-Instruct-4bit

Questions

How much VRAM does Qwen2.5 Math 7B Instruct need?

Qwen2.5 Math 7B Instruct needs 4.3 GB for its weights at 4-bit, 5.4 GB with an 4K context and 5.4 GB with 4K. At 8-bit, 8.6 GB.

What is the minimum VRAM for Qwen2.5 Math 7B Instruct?

About 4.5 GB at 3-bit with an 4K 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 Qwen2.5 Math 7B Instruct?

7.6B parameters: 14.2 GB of weights at 16-bit, 4.3 GB at 4-bit.

Can Qwen2.5 Math 7B Instruct run on an RTX 4090?

Yes. At 8-bit it holds 4K of context and writes about 87 tokens a second.

More models