TF Conceptual Parity Note (Keras Map of L1-L5, No TF Lab)

# TF Conceptual Parity Note (Keras Map of L1-L5, No TF Lab) PyTorch-only labs; this note maps the five PyTorch concepts to Keras equivalents so TF-shop interviews hold no surprises. Reading-level only. ## Mapping table | PyTorch (labs) | Keras equivalent | Note | |---|---|---| | Tensor + autograd | `tf.Tensor` + `GradientTape` | Tape scopes the watched ops | | `nn.Module` + training loop | `Model.fit` / custom `train_step` | fit hides the loop; know both | | Conv2d/MaxPool stacks | `Conv2D`/`MaxPooling2D` | Same receptive-field math | | Embedding + LSTM/GRU + masking | `Embedding(mask_zero=True)` + `LSTM/GRU` | Masking semantics match | | Freeze (`requires_grad=False`) + LR groups | `layer.trainable=False` + optimizers | Same discipline, new API | ## Recap Five rows memorized with the API asymmetry per row (Tape scoping, fit-vs-loop, mask_zero) — enough to whiteboard either framework.