VishalMysore/mayaWasm

Maya for the browser (int8 ONNX)

Maya, a 150M-parameter yes/no gate, exported for ONNX Runtime Web: one graph, int8 weight-only quantization (MatMulNBits block 128, int8 embeddings), 161 MB.

• model.onnx + model.onnx.data split into 24 MiB parts (model.onnx.data.partNNN); manifest.json lists the parts with the total size and SHA-256. Concatenate the parts in order to get model.onnx.data.
• Inputs input_ids, attention_mask (int64, [batch, seq]) from tokenizer.json, encoding the pair (text, statement); output logits [batch, 2].
• P(yes) = sigmoid((logits[:, 0] - logits[:, 1]) / temperature), temperature in maya_config.json (2.8).
• Checked against PyTorch on 256 hand-labeled answers: 81.6% accuracy (PyTorch 81.25%), 1 answer flips, largest probability change 0.12.

Code and evaluation: https://github.com/vishalmysore/maya. Same weak spots as the PyTorch model (agent guardrails, minimal pairs in unseen domains); do not use it as the only safety check.