Add CLAUDE.md with AI agent instructions and quick reference (#4195)

* Update base for Update on "Add CLAUDE.md with AI agent instructions and quick reference"


Replace empty placeholder with structured documentation for AI coding
assistants (Claude Code, Cursor, Copilot). Includes config class table,
granularity reference, deprecated API warnings, and pointers to in-repo
docs for architecture details.

Comparison: Old CLAUDE.md vs New CLAUDE.md
Instructions+Scripts for repro available in https://github.com/supriyar/torchao-eval 
Setup:
  - Subject model: Claude Sonnet 
  - Judge model: Claude Opus 
 
 Sonnet/Opus Results (61 prompts, final)

<img width="586" height="257" alt="image" src="https://github.com/user-attachments/assets/fc1ff374-eb02-40ed-91c7-089f55715144" />

  

  




[ghstack-poisoned]

Author: supriyar

CLAUDE.md

@@ -1,3 +1,51 @@
-# TorchAO Claude Instructions
+# TorchAO
 
-Fill me in
+PyTorch-native library for quantization, sparsity, and low-precision training.
+
+## Config Classes
+
+All configs inherit from `AOBaseConfig`. Defined in `torchao/quantization/quant_api.py`. Use `FqnToConfig` to apply different configs to different layers by module name.
+
+## Stable vs Prototype
+
+- **Stable** (`torchao/quantization/`, `torchao/float8/`, `torchao/sparsity/`, `torchao/optim/`): API stability guaranteed.
+- **Prototype** (`torchao/prototype/`): Experimental, API may change without notice.
+
+See [docs/source/workflows/index.md](docs/source/workflows/index.md) for the full dtype x hardware status matrix.
+
+## Architecture and Contributing
+
+- [Quantization Overview](docs/source/contributing/quantization_overview.rst) - full stack walkthrough, tensor subclasses, quantization flows
+- [Contributor Guide](docs/source/contributing/contributor_guide.rst) - how to add tensors, kernels, configs
+- [Inference Workflows](docs/source/workflows/inference.md) - which config to use for which hardware
+- [PT2E Quantization](docs/source/pt2e_quantization/index.rst) - PyTorch 2 Export quantization for deployment backends (X86, XPU, ExecuTorch)
+
+These render at https://docs.pytorch.org/ao/main/
+
+## Deprecated APIs
+
+Do not use or recommend these:
+- `AffineQuantizedTensor` (AQT) in `torchao/dtypes/` - old v1 system, being removed
+- `autoquant()` - deleted
+- Layout registration system (`PlainLayout`, `Float8Layout`, `TensorCoreTiledLayout`, etc.) - deleted
+- `TorchAODType` - deprecated
+- `change_linear_weights_to_int4_woqtensors` - deleted, use `quantize_(model, Int4WeightOnlyConfig())`
+
+New tensor types should inherit from `TorchAOBaseTensor` in `torchao/utils.py`, not AQT.
+
+## Development
+
+```bash
+# Setup
+USE_CPP=0 pip install -e . --no-build-isolation   # CPU-only
+USE_CUDA=1 pip install -e . --no-build-isolation   # With CUDA
+
+# Test (mirrors source structure)
+pytest test/quantization/test_quant_api.py
+pytest test/float8/
+pytest test/prototype/mx_formats/
+```
+
+## Commit Messages
+
+- Do not commit without explicit request from the user