implementing-llms-litgpt
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides standard technical documentation for implementing LLMs. The demonstrated workflows, including model configuration and training, follow established industry practices for the LitGPT framework.\n- [EXTERNAL_DOWNLOADS]: The skill includes instructions to install the
litgptlibrary and download model checkpoints from Hugging Face. These represent standard assets from trusted and well-known organizations in the AI research community.\n- [INDIRECT_PROMPT_INJECTION]: The skill documents workflows for fine-tuning models on external instruction datasets. While this involves processing external data, it is the primary intended function of the skill.\n - Ingestion points: Instruction-tuning datasets in JSON format (
data/my_dataset.json) and pretraining text corpora (data/my_corpus.txt).\n - Boundary markers: The Alpaca-style JSON format provides structural separation between instructions and input data.\n
- Capability inventory: The skill uses
litgptCLI tools for training, which involves GPU execution and local file system writes for model checkpoints.\n - Sanitization: Consistent with professional ML tooling, the user is responsible for the quality and safety of their training data.
Audit Metadata