GLM-4-9B-Chat
| Property | Value |
|---|---|
| Parameter Count | 9.4B parameters |
| Model Type | Transformer-based Chat Model |
| Context Length | 128K tokens |
| License | GLM-4 |
| Paper | arxiv:2406.12793 |
What is glm-4-9b-chat?
GLM-4-9B-Chat is an advanced multilingual language model developed by THUDM, representing the latest generation in the GLM-4 series. This model combines impressive performance across semantic understanding, mathematics, reasoning, coding, and knowledge tasks with practical features like multi-turn dialogue and tool integration.
Implementation Details
The model utilizes BF16 precision and supports both transformers and vLLM backends for inference. It's designed with state-of-the-art architecture enabling efficient processing of long-form content up to 128K tokens in context length.
- Supports 26 languages including English, Chinese, Japanese, Korean, and German
- Implements function calling capabilities with high accuracy (81% overall accuracy)
- Features advanced long-text processing abilities
Core Capabilities
- Achieves 72.4% on MMLU and 75.6% on C-Eval benchmarks
- Excels in coding tasks with 71.8% accuracy on HumanEval
- Superior performance in multilingual tasks across FLORES, MGSM, and other datasets
- Advanced tool integration and function calling capabilities
Frequently Asked Questions
Q: What makes this model unique?
GLM-4-9B-Chat stands out for its balanced combination of size and capabilities, offering strong performance across multiple languages while maintaining reasonable computational requirements. Its ability to handle 128K context windows and high accuracy in function calling make it particularly versatile.
Q: What are the recommended use cases?
The model is well-suited for multilingual applications, code generation, long-form content analysis, and tool integration scenarios. It's particularly effective for applications requiring both chat functionality and complex task execution like mathematical reasoning or code generation.
