The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides code snippets that download foundation models (e.g., Llama 3) and datasets from the Hugging Face Hub using official libraries like transformers and datasets. These are standard operations for the stated purpose of the skill.
[COMMAND_EXECUTION]: In the deployment optimization guide, the skill uses subprocess.run to execute external scripts and binaries required for model conversion and quantization (such as llama.cpp utilities). These commands are local to the environment and essential for preparing models for efficient inference.
[DATA_EXFILTRATION]: The training configurations mention optional integration with experiment tracking platforms like Weights & Biases (wandb) and provide methods to upload merged models back to the Hugging Face Hub. These operations involve network connectivity but are documented as intended features for model development workflows.

fine-tuning-expert