m-a-p/Infinity-Instruct-3M-0625-Qwen2-7B-COIG-P

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Apr 2, 2025Architecture:Transformer Cold

The m-a-p/Infinity-Instruct-3M-0625-Qwen2-7B-COIG-P is a 7.6 billion parameter instruction-tuned language model based on the Qwen2 architecture. It is specifically fine-tuned using the COIG-P Chinese preference dataset, which focuses on aligning with human values. This model is designed for tasks requiring high-quality, preference-aligned responses, particularly in Chinese language contexts, leveraging its substantial context length of 131072 tokens.

Loading preview...

Model Overview

The m-a-p/Infinity-Instruct-3M-0625-Qwen2-7B-COIG-P is a 7.6 billion parameter language model built upon the Qwen2 architecture. Its primary distinction lies in its fine-tuning with the COIG-P: A High-Quality and Large-Scale Chinese Preference Dataset for Alignment with Human Values. This dataset is detailed in the paper COIG-P, indicating a strong focus on generating responses that are aligned with human preferences and values, particularly within a Chinese linguistic and cultural context.

Key Characteristics

  • Architecture: Qwen2-7B base model.
  • Parameter Count: 7.6 billion parameters.
  • Context Length: Features a substantial context window of 131072 tokens, enabling processing of very long inputs and generating coherent, extended outputs.
  • Alignment: Fine-tuned with the COIG-P dataset, emphasizing alignment with human values and preferences.

Potential Use Cases

Given its specialized training, this model is likely well-suited for:

  • Applications requiring preference-aligned text generation, especially in Chinese.
  • Tasks benefiting from a large context window, such as summarization of lengthy documents or complex conversational AI.
  • Developing instruction-following agents where ethical and value-aligned responses are critical.

Further details on specific training data, evaluation metrics, and intended uses are marked as "More Information Needed" in the original model card, suggesting that users should exercise caution and conduct their own evaluations for specific applications.