clemenshemmerling/TinyGPT2-CoreML

TinyGPT2 for TinyBrain

This repository packages sshleifer/tiny-gpt2 for the TinyBrain runtime on Apple Watch.

TinyBrain bundle contract

The conversion script produces a single bundle archive:

• tinygpt2_only_logits.mlmodelc.zip

The zip layout is:

• tinygpt2_only_logits.mlmodelc/
• tokenizer/
• metadata.json

Runtime interface

• Task: causal language modeling
• Inputs:
  • input_ids (Int32, shape 1 x 16)
  • attention_mask (Int32, shape 1 x 16)
• Output:
  • logits

Files included in tokenizer/

• tokenizer.json when provided by Transformers
• vocab.json
• merges.txt
• special_tokens_map.json
• tokenizer_config.json
• token_decoder.json

Build

Run:

python3 convert_tinygpt2_to_coreml.py

The script:

1. Downloads the model and tokenizer from Hugging Face.
2. Converts the model to Core ML.
3. Compiles the package to .mlmodelc.
4. Writes TinyBrain runtime metadata.
5. Packages the compiled model, tokenizer, and metadata into one zip.