Incorrect Variance Term in Gaussian Log Probability Calculation

## Problem
The log probability calculation contains an error in the variance term:
```python
log_prob_matrix = ... - 0.5 * log_sigma.sum(dim=-1)
```
`- 0.5 * log_sigma.sum(dim=-1)` is theoretically inconsistent with Gaussian distributions

## Solution
Replace with:
```python
log_prob_matrix = ... - log_sigma.sum(dim=-1)
```