Re-sync with internal repository (#10)

The internal and external repositories are out of sync. This Pull Request attempts to brings them back in sync by patching the GitHub repository. Please carefully review this patch. You must disable ShipIt for your project in order to merge this pull request. DO NOT IMPORT this pull request. Instead, merge it directly on GitHub using the MERGE BUTTON. Re-enable ShipIt after merging.

fbshipit-source-id: 0a51b956d84f761124079edc03a7c7fc0c706c82

Co-authored-by: Facebook Community Bot <6422482+facebook-github-bot@users.noreply.github.com>

Author: meta-codesync[bot]

README.md

@@ -0,0 +1,42 @@
+# Ads Model Kernel Library
+
+High-performance GPU kernels for Meta Ads Recommendation Systems, developed by Meta Ads AI. This library provides optimized GPU kernel implementations that have been published.
+
+## Projects
+
+More coming soon!
+
+| Project | Description | Architecture | Path |
+|---------|-------------|--------------|------|
+| [GDPA](gdpa/) | Generalized Dot Product Attention kernels | Blackwell (SM100) | `gdpa/` |
+| [TLX Block Attention](block_attention/) | Triton TLX block attention kernels | Blackwell (SM100) | `block_attention/` |
+
+## Requirements
+
+- Python >= 3.10
+- PyTorch >= 2.0
+- NVIDIA GPU with Hopper (SM90) or Blackwell (SM100) architecture
+- CUDA >= 12.0
+- [nvidia-cutlass-dsl](https://github.com/NVIDIA/cutlass) >= 4.1.0
+
+## Installation
+
+```bash
+pip install nvidia-cutlass-dsl>=4.1.0 torch einops
+```
+
+## Quick Start
+
+See individual project READMEs for detailed usage:
+
+- [GDPA Quick Start](gdpa/README.md#quick-start)
+
+## Contributors
+
+**Meta Ads AI:** Jiaqi Xu, Chao Chen, Hongtao Yu, Dev Shanker, Jacky Zhou, Han Xu, Jake Siso, Xiaoyi Liu, Huayu Li, Markus Hoehnerbach, Manman Ren
+
+More contributors will be added as we publish more kernels.
+
+## License
+
+Apache License 2.0. See [LICENSE](LICENSE) for details.

block_attention/README.md

@@ -0,0 +1,26 @@
+# TLX Block Attention
+
+Triton TLX-based block attention kernels for variable-length attention workloads.
+
+## Layout
+
+- `triton/` - Triton TLX kernel implementation
+- `tests/` - project tests
+
+## Quick Start
+
+```python
+from ads_mkl.ops.oss.block_attention.triton.tlx_block_attention import (
+    block_attention_api,
+)
+```
+
+## Run
+
+```bash
+buck2 test fbcode//ads_mkl/ops/oss/block_attention:test_tlx_block_attention
+```
+
+This test target includes:
+- `fbcode/ads_mkl/ops/oss/block_attention/triton/tests/test_tlx_block_attention.py`
+- `fbcode/ads_mkl/ops/oss/block_attention/tests/block_attention_test.py`