[Bug]: tqdm progress bar seems to be wrong.

[WJMacro]

Your current environment

The output of `python collect_env.py`

Your output of `python collect_env.py` here

Model Input Dumps

No response

🐛 Describe the bug

When setting $n$ to a larger number. The tqdm progress bar shows wrong prompt number.

For example, when inferencing 500 prompts with $n=10$, tqdm shows:

Processed prompts: 10%|█▍ | 500/5000 [25:04<3:45:42, 3.01s/it, est. speed input: 36.58 toks/s, output: 4633.88 toks/s]

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[Ricardokevins]

+1

Hit the same bug

[wenlongzhao094]

Hit the same bug. Same issue as #10949 .

With v0.6.0, tqdm shows the number of prompts and a fast "est. speed input".

With v0.6.6, tqdm shows (the number of prompts x the number of generations per prompt) and a very slow "est. speed input".