Skip to content
#

greedy-decoding

Here are 9 public repositories matching this topic...

Language: All
Filter by language

A from-scratch Transformer for English-to-Chinese translation in a single PyTorch notebook. Follows the original encoder-decoder design with ~7.5M parameters, 15 epochs, and greedy decoding for short sentences. Built for readability as a teaching implementation. MIT licensed.

  • Updated Nov 16, 2026
  • Jupyter Notebook
turbo-llm

Turbo LLM: exact (token-identical) Llama 3.2 1B Instruct inference on x86 CPUs in one dependency-free C file. Lossless 12-bit weights, an exact pruned output layer and EAGLE-3 speculative decoding, reading about a third of the memory per token of a bf16 runner.

  • Updated Nov 18, 2026
  • C

Add this topic to your repo

To associate your repository with the greedy-decoding topic, visit your repo's landing page and select "manage topics."

Learn more