Zai has officially released GLM-5, a large language model (LLM) designed to tackle complex challenges in systems engineering and for tasks requiring long-term planning.
Key Features
GLM-5 represents a significant step forward compared to its predecessor, GLM-4.5, thanks to a substantial increase in the number of parameters, from 355 billion (32 billion active) to 744 billion (40 billion active). The model was pre-trained on a dataset of 28.5T tokens, compared to 23T for GLM-4.5.
Another distinctive feature of GLM-5 is the integration of DeepSeek Sparse Attention (DSA), a technique that aims to reduce deployment costs while maintaining the ability to handle large contexts. This is particularly relevant for companies wishing to run models of this type on-premise.
Resources
The model, source code, and further information are available on the following repositories:
- Blog: https://z.ai/blog/glm-5
- Hugging Face: https://huggingface.co/zai-org/GLM-5
- GitHub: https://github.com/zai-org/GLM-5
For those evaluating on-premise deployment, there are trade-offs to consider. AI-RADAR offers analytical frameworks on /llm-onpremise to evaluate these aspects.
๐ฌ Commenti (0)
๐ Accedi o registrati per commentare gli articoli.
Nessun commento ancora. Sii il primo a commentare!