Loading...
Loading...
Found 132 Skills
Systematic approach to exploring the TensorRT-LLM codebase before implementing new features or optimizations. Teaches how to discover existing infrastructure, trace code paths, and avoid reimplementing what already exists. Derived from real mistakes where ~250 lines of code were written and deleted because existing forward methods weren't discovered upfront. Use when starting any new feature, optimization, or code modification in TRT-LLM.
Techniques for reducing peak GPU memory in Megatron Bridge — expandable segments, parallelism resizing, activation recompute, CPU offloading constraints, and common OOM fixes.
Code instrumentation for timing workloads. Two scenarios: (1) Training loop — inject manual timing to report per-iteration latency, throughput (samples/sec), and data load time. (2) Standalone kernel/op — write CUDA event timing code with warmup, per-iteration statistics, and anti-pattern avoidance. Also covers NVTX annotation for labeling profiler timelines. NOT for: running or analyzing profiler tools (nsys, ncu, Nsight Systems, Nsight Compute), writing kernels (Triton, CuTe, CUDA), applying optimizations (CUDA Graphs, gradient checkpointing, fusion), or interpreting roofline/SOL% metrics. Triggers: "measure throughput", "benchmark this function", "time my training loop", "samples per second", "NVTX annotate", "instrument my dataloader", "data load time", "kernel timing", "how do I time".
Use this skill when building computer vision applications, implementing image classification, object detection, or segmentation pipelines. Triggers on image classification, object detection, YOLO, semantic segmentation, image preprocessing, data augmentation, transfer learning, CNN architectures, vision transformers, and any task requiring visual recognition or image analysis.
Validate and use MoE expert-parallel communication overlap in Megatron-Bridge, including overlap_moe_expert_parallel_comm, delay_wgrad_compute, and flex dispatcher backends such as DeepEP and HybridEP.
Image processing, object detection, segmentation, and vision models. Use for image classification, object detection, or visual analysis tasks.
Pick Jetson-compatible containers, vLLM runtime images, and Jetson AI Lab PyPI indexes; maps Orin SM 8.7 vs Thor SM 11.0 and JetPack-specific package choices.
Step-by-step tutorial for adding a heavyweight AOT CUDA/C++ kernel to sgl-kernel (including tests & benchmarks)
Create Earth2Studio diagnostic model wrappers for single-step data transformations, including simple derived diagnostics, packaged AutoModel diagnostics, and generative or diffusion diagnostics. Do NOT use for prognostic time-stepping models, data sources, or installation.
Guide for selecting and configuring distributed training strategies in NeMo AutoModel, including FSDP2, Megatron FSDP, DDP, and parallelism settings.
Add a new cuTile GPU kernel operator to TileGym. Covers dispatch registration in ops.py, cuTile backend implementation, __init__.py exports, test creation, and benchmark in tests/benchmark. Use when adding, creating, or implementing a new cuTile operator/kernel in TileGym, or when asking how to register a new cuTile op.
Use when writing Helion (import helion; import helion.language as hl) kernels over nested/ragged/jagged tensors (offsets/variable-length rows), OR when managing Helion autotuning that is slow — saving/loading tuned configs, selecting configs by GPU architecture and input shape, or avoiding re-tuning across machines. Covers hl.jagged_tile, Config.save/load, the autotune cache, helion.from_cache, AOT heuristics (aot_kernel/aot_runner), and shape specialization.