Qwen2-72B

Qwen2-72B is the flagship model of the series, featuring 72 billion parameters. Its architecture includes 80 layers with a hidden size of 8192 and implements the Grouped Query Attention mechanism with 64 query heads and 8 shared key-value heads. Combined with Dual Chunk Attention and YARN technologies, this design ensures maximum performance in processing long contexts and efficient management of KV-cache memory.

The model was trained on a high-quality dataset of 7 trillion tokens, offering maximum data diversity. The base version of the model achieves outstanding results on key benchmarks: 84.2 on MMLU, 37.9 on GPQA, 64.6 on HumanEval, 89.5 on GSM8K, and 82.4 on BBH. The instruction-tuned version, Qwen2-72B-Instruct, scores 9.1 on MT-Bench, 48.1 on Arena-Hard, and 35.7 on LiveCodeBench, placing it among the top-tier proprietary models.

Qwen2-72B demonstrates exceptional capabilities in complex reasoning, step-by-step problem solving, advanced programming, and deep contextual understanding. Its multilingual support enables professional-level performance in more than 30 languages, including Russian. Accordingly, it is designed for the most demanding AI use cases—high-level scientific research, complex software development, creation of high-quality professional content, advanced data analysis, automation of complex business processes, and intelligent decision-making systems.

Announce Date: 24.07.2024
Parameters: 72B
Context: 32K
Layers: 80
Attention Type: Full Attention
VRAM requirements: 46.0 GB using 4 bits quantization
Developer: Qwen
Transformers Version: 4.40.1
License: Tongyi-Qianwen

Use our pre-built public endpoints for free to test inference and explore Qwen2-72B capabilities. You can obtain an API access token on the token management page after registration and verification.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.