Conca979/tensor-engine-nmt ? reverse-engineered prompt

Reverse engineered prompt

Build me a from scratch English to Vietnamese translation project that works without PyTorch or TensorFlow, using NumPy with optional CuPy GPU support when available.

I want it to train a sequence to sequence model with a stacked bidirectional LSTM encoder, a stacked LSTM decoder, and Luong style attention. It should include a custom BPE tokenizer, dataset caching so tokenization is only done once, masked cross entropy, manual backpropagation, gradient clipping, Adam, checkpoint saving and resume, and beam search for inference.

Please make it usable from the command line for both training and demo translation, and include a small laptop friendly mode plus a larger GPU mode. If anything is unclear, look up current docs online if you need to, but keep the whole thing self contained and practical to run on real data.

Are you gonna build this?

make sure you review the code using coderabbit

Try freeSponsored — opens CodeRabbit in a new tab