Vijay Janapa Reddi 8cff435db9 fix(module-11): Fix Embedding and PositionalEncoding gradient flow ...

- Embedding.forward() now preserves requires_grad from weight tensor
- PositionalEncoding.forward() uses Tensor addition (x + pos) instead of .data
- Critical for transformer input embeddings to have gradients

Both changes ensure gradient flows from loss back to embedding weights

2025-10-27 20:30:03 -04:00

..

source

fix(module-11): Fix Embedding and PositionalEncoding gradient flow

2025-10-27 20:30:03 -04:00