examples/README.md catalogues 50+ runnable demos, but after the curated "Start here" block it is a long flat list. New visitors cannot easily find "the LLM ones" or "the training ones".
Task: add a grouped table (or grouped subsections) organizing the demos by domain, each row = filename + one-line blurb pulled from the script's module docstring. Suggested groups:
- LLMs / generation --
gpt_generate_ane.py, llm_chat.py, llm_prefill.py, moe_chat.py, qwen35_chat.py, spec_chat.py, llama_block_causal.py, ...
- Vision --
resnet18.py, vit.py, sd15.py, sd_unet.py, sd_vae.py, superres_espcn.py, ...
- Training on the engine -- the
train_*.py family
- Scientific / spectral --
fluid_vorticity.py, heat_equation.py, poisson_spectral.py, nbody.py, reaction_diffusion.py, eigenvalues_svd.py, spectral_analysis.py, ...
- DSP / embeddings / ONNX --
fft.py, sentence_embeddings.py, rag_embeddings.py, onnx_import.py, ...
Keep the existing "Start here" hero block at the top. Do not change any demo code -- this is purely examples/README.md (and, if it helps, a matching pass on the top-level README's examples pointer).
Why: the examples are one of the project's best on-ramps; a domain index makes them discoverable.
No Apple Silicon required -- this is a docs-only change. A reviewer with a Mac will sanity-check that each blurb matches its script's docstring.
examples/README.mdcatalogues 50+ runnable demos, but after the curated "Start here" block it is a long flat list. New visitors cannot easily find "the LLM ones" or "the training ones".Task: add a grouped table (or grouped subsections) organizing the demos by domain, each row = filename + one-line blurb pulled from the script's module docstring. Suggested groups:
gpt_generate_ane.py,llm_chat.py,llm_prefill.py,moe_chat.py,qwen35_chat.py,spec_chat.py,llama_block_causal.py, ...resnet18.py,vit.py,sd15.py,sd_unet.py,sd_vae.py,superres_espcn.py, ...train_*.pyfamilyfluid_vorticity.py,heat_equation.py,poisson_spectral.py,nbody.py,reaction_diffusion.py,eigenvalues_svd.py,spectral_analysis.py, ...fft.py,sentence_embeddings.py,rag_embeddings.py,onnx_import.py, ...Keep the existing "Start here" hero block at the top. Do not change any demo code -- this is purely
examples/README.md(and, if it helps, a matching pass on the top-level README's examples pointer).Why: the examples are one of the project's best on-ramps; a domain index makes them discoverable.
No Apple Silicon required -- this is a docs-only change. A reviewer with a Mac will sanity-check that each blurb matches its script's docstring.