The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's primary function is to submit training jobs to cloud GPUs using remote scripts. This includes instructions for running scripts directly from Hugging Face repositories and official Hugging Face TRL examples, which are considered reputable sources.
[COMMAND_EXECUTION]: The scripts/convert_to_gguf.py script uses subprocess.run to perform system-level tasks such as cloning the llama.cpp repository, installing build tools, and executing the conversion process. These commands are necessary for the model quantization workflow.
[EXTERNAL_DOWNLOADS]: The skill downloads the llama.cpp repository from GitHub and installs required Python packages from PyPI. These are standard dependencies for model conversion and training tasks.
[DYNAMIC_EXECUTION]: scripts/convert_to_gguf.py performs runtime compilation of C++ components (using CMake) to build the llama-quantize utility. This is an expected step for generating model quantizations locally in the job environment.
[DATA_EXPOSURE]: The skill handles HF_TOKEN credentials for authenticating with the Hugging Face Hub. It follows best practices by using environment variables and secret placeholders ($HF_TOKEN) to manage these sensitive tokens securely during job submission.

huggingface-llm-trainer