examples

Examples

This file is the canonical list of runnable scripts and notebooks maintained under examples/.

PyTorch + quantization demos

• examples/demo_llama_conversion.py — Convert a Hugging Face Llama checkpoint, swap every torch.nn.Linear for t81.nn.Linear, and inspect the cached ternary weights.
• examples/scaling_laws_ternary.py — Compare ternary vs float scaling laws across RMSNorm, RoPE, and throughput axes.
• examples/ternary_sparse_preview.py — Explore hybrid sparsity, GEMM packing, and quantized transformer inference with notebook-friendly visuals.
• examples/ternary_quantization_demo.ipynb — Tutorial notebook showing packed GEMMs, quantized trits, and dequantization.
• examples/ternary_phi3_ptq_qat_demo.ipynb — End-to-end Phi-3-mini PTQ/QAT notebook with size, latency, and perplexity tracking.
• examples/ternary_transformer_demo.ipynb — Micro GPT stack with cached ternary projections and packed GEMM profiling.
• examples/ternary_mnist_demo.ipynb — Quantize an MNIST classifier, pack weight buffers, and route inference through t81lib.gemm_ternary.
• examples/ternary_qat_inference_comparison.py — Run a miniature QAT loop, log ternary threshold schedules, and compare latency between torch.matmul and cached TernaryTensor.

Hardware & CLI demos

• examples/ternary_hardware_sim_demo.ipynb — Build a ternary adder, trace virtual power/latency, and compare energy vs binary hardware using t81.hardware.TernaryEmulator.
• examples/cli-examples.md — Copy/paste-ready snippets for t81-convert, t81-gguf, and t81-qat workflows.

Refer to docs/use-cases.md for details on how these examples tie into broader quantization, scaling-law, and hardware experiments, and consult the quantization workflow diagram for the PyTorch → CLI → inference path.