Fireworks.ai
Official@fw-ai · United States of America
High-performance GPU kernel optimization and synchronization for FlashInfer-based distributed computing environments.
Agent Skills by Fireworks.ai
Showing 4 vetted skills indexed across 1 GitHub repositories.
fw-airebase-upstream-release
Synchronize a GitHub fork with the latest upstream release tag via rebase branch and pull request.
fw-aibenchmark-kernel
Benchmark FlashInfer GPU kernels with CUPTI timing and CSV metrics.
fw-aidebug-cuda-crash
Log FlashInfer API inputs to diagnose CUDA crashes.
fw-aiadd-cuda-kernel
Add custom CUDA kernels to FlashInfer with C++ launchers and Python APIs.
Frequently Asked Questions About Fireworks.ai
FAQPage SchemaWhat specific tasks can be performed using these capabilities?▼
These capabilities enable precise benchmarking of FlashInfer GPU kernels, diagnostic logging for CUDA runtime crashes, and the integration of custom C++ kernels. Additionally, they facilitate the synchronization of GitHub forks with upstream release tags to ensure codebase consistency.
Which technical personas benefit from these capabilities?▼
These capabilities are designed for systems engineers, GPU performance researchers, and machine learning infrastructure developers. They are specifically intended for those working on low-level kernel optimization, distributed training performance, and maintaining custom forks of high-performance inference libraries.
What are the primary prerequisites for implementing these kernels?▼
Implementation requires a functional CUDA development environment, familiarity with C++ kernel launching patterns, and access to the FlashInfer library structure. Users must also have configured CUPTI for performance metrics and possess appropriate permissions for GitHub repository management.