Perf Torch Cuda Graph Specialist — Development agent for Claude Code
name: perf-torch-cuda-graph-specialist description: Expert in CUDA Graph capture, replay, and optimization for PyTorch.
How to install Perf Torch Cuda Graph Specialist
Installs to ~/.claude/agents/nvidia-tensorrt-llm-perf-torch-cuda-graph-specialist.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/NVIDIA/TensorRT-LLM/HEAD/.claude/agents/perf-torch-cuda-graph-specialist.md -o ~/.claude/agents/nvidia-tensorrt-llm-perf-torch-cuda-graph-specialist.md Restart Claude Code, or start a new session, for it to be picked up.
What Perf Torch Cuda Graph Specialist does
name: perf-torch-cuda-graph-specialist description: > Expert in CUDA Graph capture, replay, and optimization for PyTorch. Delegate to this agent for: (1) Analyzing code for CUDA Graph compatibility, (2) Detecting and eliminating host-device synchronizations, (3) Selecting the right CUDA Graph API (torch.compile, make_graphed_callables, TE, MCore CudaGraphManager, FullCudaGraphWrapper, manual), (4) Implementing capture and replay, (5) Full-iteration CUDA Graph capture, (6) Tro
Alternatives in Development
- Civitai Perf Review — Reviews a feature segment in the main Civitai Next.js app (src/) for production performance — N+1 queries, uni 7.2k ★
- Lateral Thinker — Subconscious subagent that surfaces cross-disciplinary structural parallels from the Vestige memory graph 632 ★
- Perf Dev — Performance Developer Agent 399 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Fk Kernel Engineer
Writes and integrates GPU kernels (Triton, TileLang, CuTe DSL, CUDA C++, CUDA graphs, torch.compile) under can
Fk Profiler
Hotspot analyst. Profiles the current candidate (torch.profiler + module attribution), explains where time goe
Kernel Cuda Specialist
PyTorch Build Resolver
PyTorch 运行时、CUDA 和训练错误解决专家。以最小变更修复张量形状不匹配、设备错误、梯度问题、DataLoader 问题和混合精度失败。当 PyTorch 训练或推理崩溃时使用。
Perf Profiling Specialist
Measure Response Time For Varying Prefix Match Lengths
import time from statistics import mean def time_compare(guess, iterations=1000): times = [] for _ in range(it
Related Skills
New Submission
Scaffold a new vortex_torch sparse-attention submission pair.
ML Production
Use when shipping ML models (classical or LLM) to production - quantization (torch.ao.quantization, AWQ/GPTQ/A
Pangu
盘古 — AI Agent 多模态记忆系统：Wing→Room→Drawer 分层存储、ONNX 本地嵌入语义检索（SQLite FTS5 + RRF 融合）、艾宾浩斯遗忘曲线与神经激活扩散；提供 REST API +
