QMoE: fail loudly when weights_prepacked=0 but PrePack did not run

[Copilot]

Description

When a QMoE model sets weights_prepacked=0 (raw [E, N, K/pack] int weights) and the session has session.disable_prepacking, PrePack() never runs, so packed_fc{1,2}_weights_ stay null and int_weights_consumed_by_prepack is false. The code then falls through to the raw initializer pointers — but those bytes are not in CUTLASS layout, so the runner consumes them as-if-prepacked and produces silently wrong output with no diagnostic.

Changes in onnxruntime/contrib_ops/cuda/moe/moe_quantization.cc (QMoE::ComputeInternal):

• Int path: Added a defensive INVALID_ARGUMENT guard — when is_int && !weights_prepacked_ but either prepack buffer is null, return a clear error instead of feeding non-CUTLASS bytes to the runner.
• wfp4afp8 native path: Same fall-through (packed_fp4_fc{1,2}_weights_ ? ... : raw) replaced with an explicit guard that errors when the repacked FP4 buffers were not produced.

Also added a focused regression test in onnxruntime/test/contrib_ops/moe_test.cc covering quant_type='int' with weights_prepacked=0 and session.disable_prepacking=1, asserting that QMoE fails with an actionable error instead of producing output.

Merged the branch with the latest main.

Motivation and Context

A prior fix removed the null-pointer crash on this path but left a misleading-success outcome that is newly user-reachable via the weights_prepacked=0 contract — the exact silent-failure mode the offline-path work set out to eliminate. These guards convert that into a loud, actionable error. The wfp4afp8 branch shares the same fall-through and is hardened for consistency.

The added regression test ensures this fail-loudly behavior remains covered going forward, especially when prepacking is disabled at the session level.

[Copilot]

Pull request overview

This PR hardens the CUDA QMoE execution path to avoid silent wrong outputs when models rely on PrePack()-produced CUTLASS layouts (int weights_prepacked=0, and native wfp4afp8) but prepacking never ran (e.g., session.disable_prepacking).

Changes:

• Add an INVALID_ARGUMENT guard for quant_type='int' && weights_prepacked=0 when the required prepacked int-weight buffers are missing.
• Add an INVALID_ARGUMENT guard for native wfp4afp8 when the repacked FP4 weight buffers are missing, instead of falling back to raw initializer bytes.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[tianleiwu]

Verdict: Approve.

This is a clean, well-scoped defensive fix that converts a silent wrong-output path into a loud, actionable error.

Correctness

• The int guard is placed right after int_weights_consumed_by_prepack is computed and fires exactly when is_int && !weights_prepacked_ but one of the prepack buffers is null — i.e. the negation of int_weights_consumed_by_prepack within that subset. After the guard, the weights_prepacked=0 int path is guaranteed to have both buffers, which also closes the partial-prepack concern noted in the surrounding comment.
• The wfp4afp8 native guard is correct: PrePackRepackFP4Weights is the only producer of packed_fp4_fc{1,2}_weights_, and the weights_prepacked attribute is int-only, so a null repacked buffer on the native path can only mean PrePack did not run. The previous fall-through to raw initializer bytes was genuinely unsafe.
• The is_packed handling on those paths keeps the source initializer alive, so failing loudly here does not risk dangling pointers.

Tests

• The new QMoETest_CUDA_Int4_DisablePrepackingFailsLoudly exercises the int branch via session.disable_prepacking=1 and asserts the actionable failure message. Because the guard returns before any kernel launch, it runs on any SM ≥ 700 and doesn't depend on CUTLASS shape constraints.

One optional maintainability suggestion left inline.