peft-fine-tuning

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). Outsider free text can enter the LLM context via the runtime dataset loading path load_dataset("databricks/databricks-dolly-15k", split="train") and subsequent dataset.map(tokenize) where example['instruction']/example['response'] (authored by third parties) are concatenated into training text.