Add Parameter to Control Input Length in run_pipeline.py for Image-to-Text Examples

[kalyank007]

I would like to request the addition of a parameter to control the input length in the run_pipeline.py script for image-to-text examples. This would allow users to specify how the input sequences are padded or truncated, making the script more flexible for different use cases.

If --max_input_tokens > 0: Pad and truncate input sequences to the specified number of tokens.
If --max_input_tokens == 0 (default): default behavior, Do not truncate input sequences and use the full input prompt.

The primary motivation for this feature request is to enable benchmarking of the run_pipeline.py script for various input and output token lengths. Currently, the script uses a fixed length for default input, limiting its flexibility for benchmarking use cases. By introducing a parameter to control the input length, users can measure the script's performance with different input lengths to better understand how token length impacts runtime, memory usage, and overall throughput.

[kalyank007]

@sywangyi, Could you please take a look at this issue. Let me know if you need any clarification or additional details. Your support on this is much appreciated!

[sywangyi]

truncate will not be used, since it may delete the image token id by mistake