Re-tune and validate GPU matmul golden configuration YAMLs via the emmy CLI.
npx skills add https://github.com/cloudrift-ai/emmy --skill tune-golden
Please help me install this Agent Skill. Skill: tune-golden Source: https://github.com/cloudrift-ai/emmy/tree/main/.claude/skills/tune-golden Command: npx skills add https://github.com/cloudrift-ai/emmy --skill tune-golden
Manually maintaining and updating per-GPU matmul golden configuration YAMLs is error-prone, time-consuming, and often leaves stale configs that underperform compared to newly tuned greedy kernel picks. This skill eliminates that manual overhead by automating full golden dataset re-tuning, live A/B validation, and auditable update reporting.
Use the tune-golden skill to re-tune all matmul golden configs for your target GPU, validate greedy pick performance against recorded goldens, and generate a findings report of updates and performance gaps.
High-intent search queries and answers about installing and using this skill.
Automating GPU matmul golden config tuning requires re-tuning the full dataset via the emmy CLI to train the learned prior across all shapes, which eliminates manual YAML maintenance errors and ensures deployed kernel configs are optimally tuned.
A/B benchmarking for GPU kernel performance validation compares deployable greedy kernel picks against recorded golden configs in a live run, identifying genuine performance wins while accounting for benchmark noise through reproducible re-runs.
Updating matmul golden YAML files without breaking formatting requires hand-safe edits that preserve required flow-style knob formatting and key order, avoiding destructive auto-serialization that would corrupt the configuration structure.
The emmy compiler supports dynamic symbolic-axis matmul goldens during re-tuning, allowing the automated workflow to handle flexible matmul shapes while applying full golden dataset re-tuning and performance validation.
Tracking prior search shortfalls in CUDA performance tuning involves generating detailed findings reports that analyze knob mismatches and workflow friction, providing actionable recommendations for improving the tuning pipeline.
The best way to validate greedy kernel picks against recorded goldens is running live A/B benchmarks within the same tuning pass, using reproducible re-runs to account for benchmark noise and confirm genuine performance improvements.
cloudrift-aiAutomate per-GPU kernel node data collection and merging for LLM compiler autotuning.
cloudrift-aiAutotune compiled LLM kernels and generate root-cause performance findings reports.
LancerLabOrchestrate GPU kernel autotuning cycles for CUDA and DSL-targeted kernels on NVIDIA GPUs.
rbutinarUpload files or folders directly to a Databricks workspace.
