Fix LiveCodeBench crashes with type safety and list conversion

[dkimds]

Summary

Fixes #136 by adding robust type checking and ensuring dataset is properly converted to list format.

Problem Solved

LiveCodeBench variants could crash or behave unexpectedly due to:

• Dataset iteration issues when not converted to list
• Missing type validation causing runtime errors

Changes

Files Modified:

• eval/chat_benchmarks/LiveCodeBench/eval_instruct.py
• eval/chat_benchmarks/LiveCodeBenchv5/eval_instruct.py
• eval/chat_benchmarks/LiveCodeBenchv5_official/eval_instruct.py

Key Improvements:

1. Explicit list conversion: examples = list(examples_dataset)
2. Type safety: Added isinstance(example, dict) checks
3. Error logging: Non-dict examples logged and skipped instead of crashing

Code Changes

# Before
examples_dataset = self.load_questions()
for idx, example in enumerate(examples_dataset):
    # Could crash if example is not dict

# After  
examples = list(self.load_questions())
for idx, example in enumerate(examples):
    if not isinstance(example, dict):
        self.logger.error(f"Example {idx} not dict: {type(example)}")
        continue
    # Safe to process

## Testing
```bash
# Test all LiveCodeBench variants work without crashes
python -m eval.eval --model hf --tasks LiveCodeBench --debug --model_args "pretrained=microsoft/DialoGPT-medium"
python -m eval.eval --model hf --tasks LiveCodeBenchv5 --debug --model_args "pretrained=microsoft/DialoGPT-medium"

Impact

✅ Prevents runtime crashes from type mismatches
✅ More robust error handling and logging
✅ Better debugging experience
✅ No functional changes to working cases

[dkimds]

Hi @neginraoof, could you please take a quick look at my PR when you have time? Thanks!