趋近智
参数
9B
上下文长度
128K
模态
Text
架构
Dense
许可证
MIT License
发布日期
30 Jun 2024
知识截止
-
注意力结构
Multi-Head Attention
隐藏维度大小
-
层数
40
注意力头
-
键值头
-
激活函数
-
归一化
RMS Normalization
位置嵌入
Absolute Position Embedding
不同量化方法和上下文大小的显存要求
The GLM-4-9B-Chat model is a conversational artificial intelligence model developed by THUDM, the Tsinghua University Department of Computer Science and Technology, in collaboration with Z.ai. It belongs to the GLM (General Language Model) series, a framework designed for broad language task performance. This model, with 4.9 billion parameters, is specifically fine-tuned for interactive dialogue and multi-turn conversational applications, excelling in the comprehension and generation of natural language.
Architecturally, GLM-4-9B-Chat is based on a dense transformer design. The underlying GLM-4 architecture incorporates key technical features such as RMSNorm for normalization and employs an autoregressive blank infilling approach during its pretraining phase. This model supports an extended context length of up to 128,000 tokens, facilitating long-range dependencies in conversational flows. The model family has enhanced its multilingual capabilities, supporting 26 languages, including Chinese, English, Japanese, Korean, and German.
GLM-4-9B-Chat is designed for a variety of general-purpose language applications. Its functional scope includes robust natural language understanding, the generation of coherent and contextually appropriate text, and proficient multilingual interaction. Beyond its core conversational abilities, the model integrates advanced functionalities such as web browsing, code execution, and the invocation of custom tools via function calls. These capabilities make it applicable for deployments in conversational AI assistants, advanced chatbots, and systems requiring automated content generation or problem-solving through external tools.
General Language Models from Z.ai
排名适用于本地LLM。
没有可用的 GLM-4-9B-Chat 评估基准。