What are the hardware requirements for Qwen2 72B?

Qwen2 72B requires approximately 145.5 GB of VRAM and supports a context window of 128K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Qwen2 72B and how large is it?

Qwen2 72B is developed by Qwen, a model with 72b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Qwen2 72B perform on standard benchmarks?

Qwen2 72B has the following published scores: MMLU 83.79. Compare against reference models on this page or on the LLM Explorer leaderboards.

Qwen2 72B by Qwen — VRAM 145.5GB, 128K context

Name: Qwen2 72B
Author: Qwen

Qwen2 72B is an open-source language model by Qwen. Features: 72b LLM, VRAM: 145.5GB, Context: 128K, License: other, LLM Explorer Score: 0.28, Arc: 68.8, HellaSwag: 87.3, MMLU: 83.8, GSM8K: 85.3.

Conversational En Endpoints compatible Eval-results Pretrained Qwen2 Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/Qwen/Qwen2-72B

Qwen2 72B Parameters and Internals

Model Type

text-generation

Additional Notes

Qwen2 includes a Mixture-of-Experts model and is based on the Transformer architecture with improvements like SwiGLU activation, attention QKV bias, group query attention, and has an adaptive tokenizer for multiple languages and code.

Supported Languages

en (English)

LLM Name	Qwen2 72B
Repository 🤗	https://huggingface.co/Qwen/Qwen2-72B
Model Size	72b
Required VRAM	145.5 GB
Updated	2026-08-04
Maintainer	Qwen
Model Type	qwen2
Model Files	3.8 GB: 1-of-37 4.0 GB: 2-of-37 3.8 GB: 3-of-37 4.0 GB: 4-of-37 4.0 GB: 5-of-37 4.0 GB: 6-of-37 3.8 GB: 7-of-37 4.0 GB: 8-of-37 4.0 GB: 9-of-37 4.0 GB: 10-of-37 3.8 GB: 11-of-37 4.0 GB: 12-of-37 4.0 GB: 13-of-37 4.0 GB: 14-of-37 3.8 GB: 15-of-37 4.0 GB: 16-of-37 4.0 GB: 17-of-37 4.0 GB: 18-of-37 3.8 GB: 19-of-37 4.0 GB: 20-of-37 4.0 GB: 21-of-37 4.0 GB: 22-of-37 3.8 GB: 23-of-37 4.0 GB: 24-of-37 4.0 GB: 25-of-37 4.0 GB: 26-of-37 3.8 GB: 27-of-37 4.0 GB: 28-of-37 4.0 GB: 29-of-37 4.0 GB: 30-of-37 3.8 GB: 31-of-37 4.0 GB: 32-of-37 4.0 GB: 33-of-37 4.0 GB: 34-of-37 3.8 GB: 35-of-37 4.0 GB: 36-of-37 3.5 GB: 37-of-37
Supported Languages	en
Model Architecture	Qwen2ForCausalLM
License	other
Context Length	131072
Model Max Length	131072
Transformers Version	4.40.1
Tokenizer Class	Qwen2Tokenizer
Padding Token	<\|endoftext\|>
Vocabulary Size	152064
Torch Data Type	bfloat16
Errors	replace

Rank the Qwen2 72B capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Quantized Models of the Qwen2 72B

Model	Likes	Downloads	VRAM
Dolphin Vision 72B 4bit	7	1038	41 GB
...in 2.9.2 Qwen2 72B 6.0bpw EXL2	1	5	56 GB
...phin 292 Qwen2 72b EXL2 4 0bpw	1	4	38 GB
...n 2.9.2 Qwen72b 8.0bpw H8 EXL2	2	3	66 GB

Best Alternatives to Qwen2 72B

Best Alternatives	Context / RAM	Downloads	Likes
...R10528DistillQwen 72B Preview3	195K / 146 GB	8	7
KAT Dev 72B Exp FP8	156K / 75.5 GB	13	6
Qwen2.5 72B	128K / 145.5 GB	13708	101
Ultiima 72B	128K / 146.1 GB	51	3
Homer V1.0 Qwen2.5 72B	128K / 146.1 GB	14	6
Ultiima 72B V1.5	128K / 146.1 GB	25	0
EVA Qwen2.5 72B V0.2	128K / 146 GB	159	22
AceInstruct 72B	128K / 146 GB	88	18
...n2.5 72B 2x Instruct TIES V1.0	128K / 146.1 GB	80	1
Dolphin 2.9.2 Qwen2 72B	128K / 146 GB	1515	162

Note: green Score (e.g. "73.2") means that the model is better than Qwen/Qwen2-72B.