from optimum.onnxruntime import ORTModelForSeq2SeqLM
from transformers import T5Tokenizer

# Load ONNX model
model = ORTModelForSeq2SeqLM.from_pretrained("Mitchins/t5-base-artgen-multi-instruct-ONNX")
tokenizer = T5Tokenizer.from_pretrained("Mitchins/t5-base-artgen-multi-instruct-ONNX")

# Example usage
text = "woman in red dress"
prompt = f"Enhance this prompt (no lora): {text}"

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=80)
result = tokenizer.decode(outputs[0], skip_special_tokens=True)

Performance

Base Model: T5-base (Google)
Training Samples: 297,282
Parameters: 222M
Format: ONNX (FP32)
Optimization: CPU inference optimized

Optimized for CPU inference with significant speedup compared to PyTorch on CPU.