SUPER-GIANT

Create/start/stop/delete a RunPod GPU Pod using ONLY GPU type `XXX` + the existing template named `GIANT-container`. Includes Spot (Secure Cloud) via interruptible Pods.

Provides guidance for writing and benchmarking optimized CUDA kernels for NVIDIA GPUs (H100, A100, T4) targeting HuggingFace diffusers and transformers libraries. Supports models like LTX-Video, Stable Diffusion, LLaMA, Mistral, and Qwen. Includes integration with HuggingFace Kernels Hub (get_kernel) for loading pre-compiled kernels. Includes benchmarking scripts to compare kernel performance against baseline implementations.

Read an arXiv paper from TeX source and write a repo-focused summary. Use when the user references a arXiv url for a paper.

Run commands or JAX scripts on the remote CUDA GPU box via ssh root@gpu-box when execution needs a GPU/CUDA backend or the user asks to run on the GPU box/remote container.

Use s5cmd for S3 listing, move, copy, and sync tasks.

Create detailed documentation in TYPST covering the architecture and experiments conduceted in the current subproject.