技术规格

系统要求

不同量化方法和上下文大小的显存要求

OLMo 3 7B Instruct

OLMo 3 7B Instruct is an instruction-tuned variant developed by the Allen Institute for AI (AI2). It is optimized for chat, tool use, multi-turn dialogue, and function calling. This model is part of the broader OLMo 3 family, which aims to enable the scientific study of language models through complete transparency.

This model employs a Transformer-style autoregressive architecture with 7 billion parameters. Its post-training methodology includes supervised fine-tuning (SFT), Direct Preference Optimization (DPO), and Reinforcement Learning from Verifiable Rewards (RLVR) on the Dolci datasets. The base OLMo 3 models are initially trained on the Dolma 3 dataset, which integrates diverse text modalities such as web content, scientific publications, and code repositories, supporting a context length of up to 65,536 tokens.

OLMo 3 7B Instruct demonstrates capabilities in instruction-following, coding, and knowledge-based tasks. It is suitable for applications such as conversational AI systems, code development tools, and educational platforms. A defining aspect of the OLMo project is its transparency, offering public access to training data, model code, intermediate checkpoints, and training logs, which facilitates research and further development by the AI community.

没有可用的 OLMo 3 7B Instruct 评估基准。