Qwen2.5-32B

Qwen2.5-32B features 32 billion parameters, 64 layers, and a 40/8 attention head architecture, representing a significant leap in computational power and model capabilities. With support for a 128K-token context window and 8K-token generation capacity, the model can handle exceptionally complex and large-scale tasks.

Qwen2.5-32B reintroduces the 32B parameter size to the Qwen series after its absence in Qwen2, offering users a powerful alternative to the flagship 72B model with lower resource requirements. Trained on 18 trillion high-quality tokens, the model demonstrates robust performance with large datasets, expert-level knowledge in specialized domains, superior abstract reasoning capabilities, and the ability to solve problems requiring deep contextual understanding and multi-step analysis.

Qwen2.5-32B is designed for organizations and research teams that need frontier-model capabilities without the full cost of the largest models. Ideal applications include scientific research, complex software development, high-quality content creation, expert support systems in medicine and law, and as a foundation for building highly specialized AI systems.

Announce Date: 19.09.2024
Parameters: 32B
Context: 131K
Attention Type: Full Attention
VRAM requirements: 46.9 GB using 4 bits quantization
Developer: Alibaba
Transformers Version: 4.43.1
License: Apache 2.0

Use our pre-built public endpoints to test inference and explore Qwen2.5-32B capabilities.

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.