Supports batching in vLLM offline inference script

[learning-to-play]

Description

Support batching in vLLM offline inference script

This change enables batching in the vllm_decode.py offline inference script by duplicating the input prompt based on the per_device_batch_size configuration parameter.

Details:

• Replaces the hardcoded single-prompt list with a list multiplied by per_device_batch_size.
• Explicitly casts per_device_batch_size to int to prevent TypeError when the parameter defaults to a float (e.g., 12.0 in base.yml).
• Allows users to benchmark vLLM offline inference throughput with larger batch sizes by overriding per_device_batch_size on the command line.

Tests

I verified this change by running the vllm_decode script on a TPU v5e-8 VM with per_device_batch_size set to different values, confirming that it batches correctly and generates outputs.

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!