Arcaine

Query available hardware and obtain details about the device architecture to inform hardware aware implementations.

explore repositories under /reference/{repository_name}

Analyze a Hugging Face transformers model and implement its inference pipeline in SYCL. Use this skill whenever the task involves porting, implementing, or optimizing a transformers architecture (LLM, VLM, encoder, or any modality) as SYCL kernels or a SYCL runtime — including when the user mentions oneAPI, DPC++, Intel GPUs, Level Zero, a "custom inference engine", or asks to run a Hugging Face checkpoint without Python. Also use it at the start of such tasks to query the available hardware and device architecture so the implementation is hardware-aware from the beginning

Use `docker exec` to run commands in the specified container.