ssadqdsacf/cross-unlearning-case6-qwen35-4b-npo-oracle1-epoch10
The ssadqdsacf/cross-unlearning-case6-qwen35-4b-npo-oracle1-epoch10 is a 4.5 billion parameter language model with a 32768 token context length, developed by ssadqdsacf. It is an NPO unlearning baseline model, initialized from a selected epoch-10 SFT model. This model is specifically designed for cross-unlearning tasks, focusing on classification, masking, and generation, and is part of a formal matrix of six models.
Loading preview...
Model Overview
The ssadqdsacf/cross-unlearning-case6-qwen35-4b-npo-oracle1-epoch10 is a 4.5 billion parameter language model with a 32768 token context length, developed by ssadqdsacf. It serves as an NPO (Negative Preference Optimization) unlearning baseline, initialized from a specific epoch-10 SFT (Supervised Fine-Tuning) model. This model is one of six formal matrix models, focusing on the challenging task of "unlearning" specific information or behaviors.
Key Capabilities
- Cross-Unlearning Baseline: Provides a foundational model for research and application in unlearning specific data or patterns.
- Multi-directional Evaluation: Includes verified capabilities for three-direction classification, masking, and generation tasks, crucial for assessing unlearning effectiveness.
- Detailed Provenance: Offers extensive machine-readable provenance information, including protocol, SHA256 hashes, model fingerprint, and training run manifests, ensuring transparency and reproducibility.
Good For
- Research in Model Unlearning: Ideal for researchers and developers exploring methods to remove or mitigate specific knowledge from large language models.
- Evaluating Unlearning Techniques: Useful for benchmarking and comparing different unlearning strategies across classification, masking, and generation tasks.
- Reproducible AI Experiments: The detailed provenance data makes it suitable for experiments requiring high levels of transparency and reproducibility in model development.