[Jiajia Qin (qjia7)]

Description

Support empty packed weights in CPU and WebGPU MatMulNBits. With weight_prepacked=0, one or more dimensions of B's packed shape may be zero; the other dimensions must retain their normal values.

The output keeps the logical shape determined by A, K and N, and contains zeros or the broadcast bias. Unused quantization parameters may also be empty. CPU skips unused parameter packing and matrix multiplication; WebGPU records a fill dispatch so graph replay rewrites the output correctly. QKV/MLP fusion remains enabled: when a projection has empty weights, the fused operator uses the shared zero/bias fallback while computing the nonempty projections.

Includes operator documentation and focused tests for FP16/FP32, 2/4/8-bit weights, constant/runtime inputs, bias, malformed shapes, empty/nonempty overrides, and graph capture.

Motivation and Context

Optional projections, such as disabled LoRA adapters, should not require full-size zero weight tensors. Zero-sized defaults let the same raw MatMulNBits graph run on CPU and WebGPU, with nonempty weights supplied when needed.

Split from #33370. That PR retains these changes until this PR merges; QLoRA-specific operators, fusion and shaders remain there.

Validation

Built the standalone branch against main with built-in WebGPU on Windows x64. All 16 empty-weight tests, 28 existing QKV/MLP fusion regressions, and 8 existing nonempty-weight regressions passed; 4 Arm64-specific tests were skipped. Coverage includes fused empty/nonempty overrides, MLP bias, residual outputs, and capture replay. FP16 WebGPU cases require confirmed ShaderF16 support. Lint and git diff --check passed.

[Copilot]

🟡 Changes recommended

Existing WebGPU fusions bypass the empty-weight handling, and the new FP16 tests need device-feature guards.

2 open issues

• Exclude potentially empty weights from WebGPU matmul fusions · New
• Gate WebGPU FP16 tests on ShaderF16 support · New

What changed in this PR

Adds empty packed-weight support to CPU and WebGPU MatMulNBits, allowing optional projections to use zero-sized defaults while preserving logical output shapes.

Changes:

• Validates empty packed shapes and returns zeros or broadcast bias without processing unused quantization data.
• Records WebGPU fill dispatches for graph replay and skips unused CPU prepacking.
• Documents the behavior and adds tests for empty weights, runtime overrides, malformed shapes, and graph capture.

File	Description
onnxruntime/​test/​contrib_ops/​matmul_nbits_empty_weights_test.cc	Adds CPU/WebGPU correctness, override, and replay tests.
onnxruntime/​core/​graph/​contrib_ops/​contrib_defs.cc	Documents empty-weight semantics in the operator schema.
onnxruntime/​contrib_ops/​webgpu/​quantization/​matmul_nbits.cc	Adds a replayable zero/bias fill path.
onnxruntime/​contrib_ops/​cpu/​quantization/​matmul_nbits.cc	Skips unused prepacking and fills empty-weight outputs.
onnxruntime/​contrib_ops/​cpu/​quantization/​matmul_nbits_helper.h	Adds shared empty-weight shape validation.
docs/​ContribOperators.md	Updates the generated operator documentation.

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

Cross-provider prepacking, fused execution, and graph-capture changes warrant final human review of native-platform validation.

No open issues 🎉

2 resolved since last review

• Exclude potentially empty weights from WebGPU matmul fusions
• Gate WebGPU FP16 tests on ShaderF16 support

_{Review effort: Balanced}

[Xiaofei Han (xiaofeihan1)]

LGTM. One non-blocking question about model portability below.