Phase 3 · The Metal

The Compiler Stack

torch.compile is load-bearing in vLLM V1 — stop treating it as magic.

0/3 · 240 XP

You’re previewing the curriculum. Sign in (top right) to track progress, take knowledge checks, and unlock the verifiers.

Tasks

Learn torch.compile for inference: Dynamo capture, static KV cache, recompile pitfalls
Compile a decode loop with a static KV cache and measure the speedup; then trigger a recompile on purpose (dynamic shape) and watch it in the logs.
read+70 XPresource ↗
Read the Triton that Inductor writes for you
TORCH_LOGS=output_code on a small fused op. Find the fusion decisions, the tiling, the pointer math — compare with the kernels you'll write in Phase 4.
build+100 XP
Read vLLM's torch.compile integration + piecewise CUDA graphs design docs
Piecewise compilation split at attention ops, captured into CUDA graphs — the concrete architecture your Phase 6 deployment runs on.
read+70 XPresource ↗