Repro opt vits

opt.py

# 1. conda create -n ryzen101 python=3.9
# 2. install Ryzen AI Software following https://ryzenai.docs.amd.com/en/latest/manual_installation.html
# 3. Run .\transformers\setup.bat
# 4. Run .\transformers\opt-onnx\setup.bat recommended in the README can not be run - the file does not exist.
# 5. Run .\set_opt_onnx_env.bat opt-125m
# 6. Run .\prepare_model.bat opt-125m
# 7. And then run:

import onnxruntime
import numpy as np
import os

model_path = r"C:\Users\Felix\HF\RyzenAI-SW\example\transformers\opt-onnx\opt-125m_ortquantized\model_quantized.onnx"

providers = ["VitisAIExecutionProvider"]
vaip_config_path = r"C:\Users\Felix\HF\RyzenAI-SW\example\transformers\opt-onnx\vaip_config.json"
cache_dir = os.path.join(os.path.dirname(model_path), "vaie_cache")

provider_options = [{
    'config_file': vaip_config_path,
    'cacheDir': str(cache_dir),
}]

print("----- LOAD MODEL")
session = onnxruntime.InferenceSession(
    model_path,
    providers=providers,
    provider_options=provider_options
)

inp = {
    "input_ids": np.random.randint(0, 5, size=(2, 5)).astype(np.int64),
    "attention_mask": np.ones((2, 5)).astype(np.int64),
}

for i in range(12):
    dummy_key = np.random.rand(2, 12, 0, 64).astype(np.float32)
    inp[f"past_key_values.{i}.key"] = dummy_key
    inp[f"past_key_values.{i}.value"] = dummy_key


print("----- RUN INFERENCE")
outputs = session.run(None, inp)

print("outputs:", len(outputs))