Fix gpt-oss model export

[apsonawane]

This pull request updates the logic for handling the block_size attribute in QMoE (Quantized Mixture of Experts) model building and quantization. The changes ensure that block-wise quantization is only used when explicitly specified, defaulting to tensor-level quantization otherwise. The most important changes are:

Quantization logic updates:

• In make_qmoe_weights, block-wise quantization is now only used if int4_block_size is explicitly present in extra_options; otherwise, tensor-level quantization is used by default. The block_size attribute in moe_attrs is set accordingly.

Operator construction improvements:

• In make_qmoe_op, the block_size attribute is only included in the operator's attributes if it was explicitly set in moe_attrs, preventing unnecessary or default values from being passed.
• The direct passing of block_size as a parameter to make_node is removed; it is now only included via extra_kwargs when appropriate.

[Copilot]

Pull Request Overview

This PR updates the QMoE (Quantized Mixture of Experts) quantization logic to distinguish between block-wise and tensor-level quantization based on whether the int4_block_size parameter is explicitly specified by the user. The key change is making block-wise quantization opt-in rather than automatic.

• Switches QMoE quantization from automatic block-size detection to explicit opt-in behavior
• Defaults to tensor-level quantization (using TensorRT-LLM) when int4_block_size is not specified
• Conditionally includes the block_size attribute in the QMoE operator based on the quantization method used

[LorenRd]

Hi @apsonawane @kunal-vaishnavi not sure if it's part of the scope of this PR but after building from this branch, exporting with DML int4 I tried to run it but it fails

onnxruntime-directml                     1.23.0
onnxruntime-genai-directml               0.11.0.dev0 C:\Users\ailab\Desktop\onnxruntime-genai\build\Windows\Release\wheel

RuntimeError: Load model from C:\Users\ailab\Desktop\xllm_lib\artifacts\models\gpt_oss_20b_onnx_dml_int4\model.onnx failed:Type Error: Type parameter (T) of Optype (SkipSimplifiedLayerNormalization) bound to different types (tensor(float16) and tensor(float) in node (/model/layers.1/input_layernorm/SkipLayerNorm).

[tianleiwu]

Please merge main to resolve conflicts.

[apsonawane]

@LorenRd sorry for the late reply. I updated the exception since dml does not support block-wise quant, earlier we were checking for cpu specifically so this PR should not affect dml export. Were you able to run it earlier?