| name | compileiq-booster-pack |
| description | Use BEFORE running a full CompileIQ search. Walks through downloading a Booster Pack from NVIDIA/CompileIQ GitHub Releases, applying ACF candidates one at a time to the user's compiler (raw PTXAS, NVCC, Triton, Helion, FlashInfer), and keeping only candidates that compile, pass correctness, and beat the no-ACF baseline. Includes the mandatory Debug-pack O0/O3 ACF-injection canary that proves the ACF is reaching PTXAS. Triggers on "booster pack", "ACF", "apply-controls", "speed up without searching", "helion fp8", "flashinfer batch decode", "debug pack".
|
| when_to_use | - Workload is close to a published pack (Helion FP8 quant / causal
depthwise conv / Gated DeltaNet fwd; FlashInfer BatchDecode is a known
related workload).
- User wants a fast shortcut before paying for a full CompileIQ search.
- User asks "is there a known-good config for X?"
Don't use when:
- User lacks a stable baseline, correctness check, or benchmark setup.
- Workload has nothing in common with available packs; skip to
compileiq-run-search.
|
| license | Apache-2.0 |
| metadata | {"version":"1.0.0","author":"NVIDIA CompileIQ","domain":"compiler-optimization"} |
| allowed-tools | Bash Read |
| paths | ["**/*.cu","**/*.cuh","**/*.acf","**/*.py","**/*.sh"] |
compileiq-booster-pack
Try curated .acf candidates before running a full CompileIQ search.
A Booster Pack is a zip of ACFs that NVIDIA validated against a specific
workload family. They are not guaranteed speedups; treat every candidate
as workload-specific and validate it on your own benchmark.
Authoritative narrative: docs/booster_packs.md, docs/flashinfer_booster.md.
When
| If this is true | Use this path |
|---|
| Workload is close to a Booster Pack's intended workload, compiler, GPU, and validation context. | Try the Booster Pack first. |
| Workload differs materially or no pack candidate helps. | Run a full CompileIQ search (compileiq-run-search). |
| Baseline, correctness check, compiler path, or benchmark setup are not in place. | Wait. Fix those before applying any ACF. |
Available packs (today)
| Pack | Workloads it was validated against | Notes |
|---|
booster-pack-helion.zip | Helion FP8 Quantization, Causal Depthwise Convolution, Gated DeltaNet Forward | Has shown benefit on FlashInfer BatchDecodeWithPagedKVCacheWrapper; related attention workloads worth testing. |
booster-pack-debug.zip | Diagnostic ACFs (O0, O3, others that disable or alter selected optimizations) | Not for speed; for debugging. Use the O0/O3 canary below before trusting any other pack. |
The public release shape is documented in docs/booster_packs.md. Read each
candidate's compiler_stages from its pack manifest and use every listed stage.
There is no runtime download API today. Don't invent one.
Steps
0. Pre-flight: the O0/O3 ACF-injection canary (mandatory first step)
The most common silent failure when applying ACFs is a framework cache (Triton,
Helion, FlashInfer's flashinfer_cubin/flashinfer_jit_cache, NVCC build
cache) serving a stale binary that ignored the ACF. The Debug pack has two
ACFs with predictable, opposite-direction signatures:
ptxas_opt0.acf: forces unoptimized PTXAS compilation. Applied → expect a (often 2-10x slower) vs. baseline.
