run2-pytorch-tensor-ops

name	run2_pytorch-tensor-ops
description	Robust PyTorch tensor operations for implementing loss functions, including handling device placement and numerical stability.

Tensor Operations

• torch.gather: Useful for selecting specific log probabilities from logits: torch.gather(log_probs, dim=-1, index=labels.unsqueeze(-1)).squeeze(-1).
• F.logsigmoid: Stable computation of log(sigmoid(x)).
• .to(device): Ensure all input tensors and the trainer components are on the same device.

---