Add decode check for Tunix SFT colab

[hengtaoguo]

Description

Add a decode function to vllm_decode.py and its usage in Tunix SFT colab. We reuse the in-memory weights from trainer.model and apply vllm inference framework to check the decoding results.

Tests

Results from a 10-step Tunix SFT on llama3.1-8b:

Adding requests: 100%|██████████| 1/1 [00:00<00:00, 1084.36it/s]
Processed prompts: 100%|██████████| 1/1 [05:27<00:00, 327.11s/it, est. speed input: 0.14 toks/s, output: 0.20 toks/s]
Generated text:


1. Big Ben - The iconic clock tower of the Houses of Parliament, located at the north end of Westminster Bridge. 2. Buckingham Palace - The official residence of the British monarch, located in the heart of London's West End. 3. Tower Bridge - A famous landmark bridge that spans the River Thames

Checklist

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

• I have performed a self-review of my code.
• 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.