[iPrakharV]

What changed

• replaced the checked-in text artifact with tiny-text-mps-v2, a compact MLP over word and character n-gram features
• kept worker inference dependency-free by exporting JSON weights and running the MLP in pure Python
• preserved compatibility with the old linear artifact format
• refreshed MPS training output, evaluation output, model card, report, and README numbers
• added tests for the v2 artifact schema and loader

Training and evaluation

• device: mps
• train time: 0.3991 s
• parameters: 2,273
• challenge evaluation: 16/16 correct

Validation

• python training/train_text_model.py
• python training/evaluate_text_model.py
• python -m ruff check .
• python -m compileall -q gateway worker training bench scripts tests
• python -m pytest -q
• python scripts/smoke_check.py against locally started gateway and worker
• python bench/run_scenarios.py --output-json /tmp/modelmesh-bench.json --output-md /tmp/modelmesh-bench.md