A portion of the training data for the DistilQwen2.5 model, released by Alibaba-PAI to help mitigate catastrophic forgetting during fine-tuning. The dataset covers domains such as mathematics, coding, and knowledge-based Q&A. It was last updated on HuggingFace on May 24, 2025.
Use Cases
- Fine-tuning language models for mathematics tasks based on the described mathematics domain content.
- Fine-tuning language models for code generation based on the described coding domain content.
- Fine-tuning language models for knowledge-based question answering based on the described Q&A domain content.
- Conducting multi-task fine-tuning experiments using the described multi-domain foundation.
Strengths
- Data is sourced from the original training set of the DistilQwen2.5 model.
- Released specifically to address the technical challenge of catastrophic forgetting.
- Covers multiple domains, including mathematics, coding, and knowledge-based Q&A.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count, file formats, and license are unknown, which may limit suitability assessment.
Provenance
- Source
- Alibaba-PAI
- Collection Method
- A portion of the dataset used for training the DistilQwen2.5 model.
- Freshness
- Last updated 2025-05-24 09:42:08; freshness should be verified.