qwen2 onnx model attention_mask && output_past_kv shape is wrong

[AACengineer]

Who can help?

using optimum exporter to export qwen2-0.5b to onnx model, I find onnx model attention_mask with shape [batch_size, past_sequence_length + 1]. And in validation, onnx inputs/outputs with shape:
onnx_inputs["input_ids"].shape = (2, 16)
onnx_inputs["attention_mask"].shape = (2, 32)
onnx_inputs["position_ids"].shape = (2, 16)
onnx_inputs["past_key_values.0.key"].shape = (2, 2, 16, 64)

onnx_output["logits"].shape=(2, 16, 151936)
onnx_output["present.0.key"].shape=(2, 2, 32, 64)

attention_mask shape should be [batch_size, sequence_length + past_sequence_length]? output_past_kv shape be [batch_size, kv_heads, sequence_length + past_sequence_length, head_dims] ? @michaelbenayoun

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction (minimal, reproducible, runnable)

optimum-cli export onnx --model qwen/qwen2-0___5b-instruct/ ./onnx --task text-generation-with-past

Expected behavior

Confirm whether there are any errors in shape