Seq2Seq model implemented with pytorch, using temperature sampling, top-k sampling, top-p sampling / nucleus sampling, and beam search.
-
Updated
Jul 30, 2022 - Python
Seq2Seq model implemented with pytorch, using temperature sampling, top-k sampling, top-p sampling / nucleus sampling, and beam search.
A character-level LSTM in PyTorch trained on ~735k characters of Journey to the West. 4,198 character tokens, 3-layer LSTM with ~9M parameters, temperature sampling, resumable training, and DataParallel-compatible checkpoints—all in one notebook. For learning and research.
NLP homework project covering AG News classification with BPE, Word2Vec and MLP, plus character-level LSTM generation of dinosaur names.
Minimal Python notebooks that explain the core technologies behind modern LLMs, one concept at a time.
A decoder-only Transformer implemented from scratch in PyTorch for character-level name generation.
To associate your repository with the temperature-sampling topic, visit your repo's landing page and select "manage topics."