趋近智
活跃参数
21B
上下文长度
128K
模态
Text
架构
Mixture of Experts (MoE)
许可证
Apache 2.0
发布日期
5 Aug 2025
知识截止
Jun 2024
专家参数总数
3.6B
专家数量
32
活跃专家
4
注意力结构
Multi-Head Attention
隐藏维度大小
-
层数
24
注意力头
64
键值头
8
激活函数
SwigLU
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
GPT-OSS 20B is a text-based language model developed by OpenAI, engineered for efficient operation on consumer-grade hardware, including desktops and laptops with constrained memory resources. It functions as a versatile instrument for a range of natural language processing tasks, with a particular emphasis on capabilities requiring robust reasoning and integration with external tools. This model is a component of the broader GPT-OSS family, which aims to provide powerful AI capabilities in an accessible and deployable format, facilitating both local and enterprise-specific applications.
排名适用于本地LLM。
排名
#16
基准 | 分数 | 排名 |
---|---|---|
StackUnseen ProLLM Stack Unseen | 0.8 | 🥈 2 |