[Feature] Support G-Pass@k and LiveMathBench

[jnanliu]

Motivation

Support the evaluation using G-Pass@k metric and update the configurations of LiveMathBench.

Modification

• Implement GPassKEvaluator, which supports all reasoning tasks through the abstract preprocess, group, and reduce function.
• Support to load LiveMathBench from huggingface.
• Implement LiveMathBenchEvaluator which is inherited from GPassKEvaluator for mathematical reasoning tasks and support restart eval from the checkpoint.
• Modify turbomind_with_tf_above_v4_33.py to ensure that gen_cfg of model_cfg is passed into the lmdeploy pipeline.
• Print gen_config of inference backend for ease of debugging.
• Modify openai_api.py to print error url for ease of debugging.

Checklist

Before PR:

• Pre-commit or other linting tools are used to fix the potential lint issues.
• Bug fixes are fully covered by unit tests, the case that causes the bug should be added in the unit tests.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, like docstring or example tutorials.

[tonysy]

Does this modification will introduce BC? @MaiziXiao

[MaiziXiao]

The old implementation will overwrite top_k and temperature if 'do_sample','top_k' and 'temperature' are all set. This change will make sure 'top_k' and 'temperature' will not be overwritten.

@jnanliu We should remove these lines here

[tonysy]

LGTM