趋近智
参数
6B
上下文长度
32.768K
模态
Text
架构
Dense
许可证
ChatGLM3-6B Model License
发布日期
27 Oct 2023
知识截止
-
注意力结构
Multi-Head Attention
隐藏维度大小
-
层数
-
注意力头
-
键值头
-
激活函数
-
归一化
-
位置嵌入
Absolute Position Embedding
不同量化方法和上下文大小的显存要求
ChatGLM3-6B-32K is an advanced large language model developed jointly by Zhipu AI and Tsinghua University's KEG Lab. This variant builds upon the foundation of ChatGLM3-6B, specifically enhancing its capabilities for processing and understanding long textual contexts. The model is engineered to effectively manage input sequences up to 32,768 tokens in length, a significant extension compared to the 8,000-token context of its predecessor.
The architectural design of ChatGLM3-6B-32K is based on the transformer framework, a common paradigm in large language models. Key innovations in this variant include updated position encoding mechanisms and a specialized training methodology tailored for long-text scenarios. This targeted approach during the conversation stage, utilizing the full 32K context length, allows the model to maintain coherence and accuracy over extended dialogues and documents.
ChatGLM3-6B-32K is designed for applications requiring deep understanding and generation of human-like text across extensive content. It natively supports various complex functionalities such as tool invocation (Function Call), code execution (Code Interpreter), and Agent tasks. This versatility makes it suitable for diverse use cases including long-form conversations, comprehensive text analysis of articles or documents, and the generation of detailed content based on provided prompts.
ChatGLM series models from Z.ai, based on GLM architecture.
排名适用于本地LLM。
没有可用的 ChatGLM3-6B-32K 评估基准。