Agentic Safety Gguf: Training and Evaluation Datasets for AI Security

80,992 examples (80,851 after deduplication) comprise the complete training and evaluation datasets for the agentic-safety-gguf model. The dataset was created by author guerilla7 and was last updated on 2026-01-06. It supports iterative continuation training methodology (V2→V3→V4) for full reproducibility of the specialized Llama 3.1 8B model for agentic AI security analysis.

Use Cases

• Training specialized AI safety models based on the described iterative continuation methodology.
• Evaluating the performance of agentic AI security analysis systems.
• Reproducing the training pipeline for the referenced Llama 3.1 8B model.
• Benchmarking AI safety techniques against a dedicated dataset.

Strengths

• Contains 80,992 total examples, providing a substantial base for model development.
• Explicitly supports an iterative continuation training methodology (V2→V3→V4) for full reproducibility.
• Designed as a complete set for both training and evaluation of a specialized AI safety model.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-01-06 03:26:01; freshness should be verified.

Provenance

Source

guerilla7 on Hugging Face

Collection Method

Likely created for training the agentic-safety-gguf model, as referenced in the associated paper.

Freshness

2026-01-06