Qwen2-0.5B

Qwen2-0.5B is an ultra-compact language model with 0.5 billion parameters, specifically designed for deployment on mobile and IoT devices. The model utilizes GQA (Grouped Query Attention) and tied embeddings to optimize performance, an architectural feature that significantly reduces energy consumption and memory usage during inference.  

Trained on a high-quality multilingual dataset of 12 trillion tokens, the model is capable of handling around 30 languages, including Russian and several relatively rare languages. Despite its compact size, it demonstrates strong performance in basic language tasks. However, the key advantage of Qwen2-0.5B is its ability to be efficiently deployed on smartphones, headphones, smart glasses, and other embedded systems.  

Its low memory and computational requirements make it ideal for edge computing applications. Qwen2-0.5B is particularly well-suited for developing personal assistants on mobile devices, simple chatbots, real-time text processing on IoT devices, and as a base model for specialized fine-tuning in resource-constrained environments.

Announce Date: 24.07.2024
Parameters: 0.5B
Context: 32K
Attention Type: Full Attention
VRAM requirements: 0.6 GB using 4 bits quantization
Developer: Alibaba
Transformers Version: 4.40.1
License: Apache 2.0

Use our pre-built public endpoints to test inference and explore Qwen2-0.5B 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.