HomerGPT2
First model for storytelling.
HomerGPT2 is a GPT-2 model fine-tuned on short stories. The original PyTorch and Flax checkpoints remain at the repository root.
ONNX INT4 variant
The onnx-int4/ folder contains the compact CPU model currently used by the
Homer demo. It includes the ONNX graph, external weight data, tokenizer, model
configuration, an artifact-specific README, and SHA-256 checksums.
- Source revision:
ae6018ad39a077624f056e420eca7e44541a0832 - Quantization: symmetric weight-only RTN INT4, block size 128
- Coverage: 97 constant-weight matrix operations plus token and position embeddings
- ONNX opset: 21
- Artifact size: approximately 223 MB
The prompt format is:
<|Title|> A story title. <|Step 1|>
See onnx-int4/README.md for runtime details, sampling settings, validation,
and limitations.