[Skill] docker-clean-build: Build clean Docker images with correct backend configuration

[sunway513]

Skill

docker-clean-build

Priority: P1 — Critical for reproducible deployment

Motivation

Building clean ATOM Docker images is error-prone. Past issues include: gfx950 detection failures causing wrong kernel selection, GEMM defaulting to torch instead of hipBLASLt (570x slowdown), missing CK-free fallback paths, and LDS constraint violations. Each of these bugs took days to diagnose. A skill that codifies all known pitfalls into a validated build process would prevent these regressions. Convention: image tag follows username_date_rocmversion_aiterversion_atomversion pattern.

What This Skill Should Do

1. Detect target architecture — Auto-detect gfx942 vs gfx950 from the build host or accept it as a parameter. Set correct dtypes.fp8 variant (e4m3fnuz for gfx942, e4m3fn for gfx950).
2. Configure GEMM defaults — Ensure tuned_gemm.py defaults to hipblaslt (not torch). Validate that the tuned GEMM CSV is present and covers the target model's shapes.
3. Set up fallback paths — Configure CK-free mode correctly: use_triton_gemm() must return True when ATOM_CK_FREE=1, matching the pattern in attention_mla.py and moe.py. Verify ASM GEMM is NOT used on gfx950 (produces garbage).
4. Validate backend paths — Run a minimal forward pass through each backend (Triton GEMM, hipBLASLt, MoE Triton) and check cosine similarity > 0.999 against FP16 reference.
5. Handle AITER build — Support both full build (with CK) and CK-free build. For CK-free: ensure jit/core.py raises RuntimeError (not SystemExit) on JIT failure.
6. Tag and verify — Tag image per convention, run smoke test (10 tokens generation), report image size and build time.

Acceptance Criteria

• Produces a working Docker image that generates coherent text
• gfx942 and gfx950 both produce correct output (cosine > 0.999 vs reference)
• GEMM defaults to hipBLASLt, not torch
• CK-free mode works when ATOM_CK_FREE=1
• ASM GEMM is disabled on gfx950 or produces correct output
• Build time is documented and image is tagged per convention