Accuracy Evals

Benchmark results showing model quality versus throughput trade-offs across different chips, quantization levels, and inference configurations.