How mayadispeler/morsent-128-mini works
Project type and purpose
• Morsent-128T is a tiny, fine-tuned GPT-2 language model packaged for inference. It’s not a multi-service web app but an ML artifact intended for text-to-Morse, Morse-to-text, and TARS-style humor within 128 tokens.
File structure
• Top-level:
• README.md – project overview and usage
• .gitattributes – VCS settings
• morsent-model/ – the model distribution package
• morsent-model/ contents:
• config.json – model hyperparameters and architecture spec
• generation_config.json – default generation settings (e.g., max_length, temperature)
• merges.txt, vocab.json – tokenizer vocabulary files (BPE merges and vocab)
• tokenizer.json, tokenizer_config.json – tokenizer implementation metadata
• special_tokens_map.json – mapping of special tokens (e.g., [/INST])
• model.safetensors – the fine-tuned model weights
README for additional insights
• Purpose: inference only, demo of a mini LLM on Morse translation and humor
• Dependencies: Hugging Face’s transformers library, PyTorch under the hood
• Usage:
• Load model & tokenizer via from_pretrained
• Tokenize prompt → run model.generate → decode output