PyTorch open-sources TorchTLX for custom kernel compilation

PyTorch's latest commit (trunk/0924527) open-sources TorchTLX, a domain-specific language (DSL) for Inductor, PyTorch's just-in-time compiler. TorchTLX was originally housed in an internal `fb/` folder within Inductor, with only minimal hooks visible to open-source users. Now, with the availability of the open-source fbtriton backend and the growing adoption of non-Triton DSLs as Inductor backends, Meta has decided to make TorchTLX fully public. The move aims to streamline future development directly in OSS fbtriton, reducing fragmentation and enabling the broader community to contribute.

For developers, this means easier access to a powerful tool for custom kernel generation. TorchTLX allows writing high-performance kernels that integrate seamlessly with PyTorch's graph compilation pipeline. By shifting to an open-source model, Meta lowers the barrier for research and experimentation in kernel DSLs, potentially accelerating performance optimizations for LLMs and other compute-heavy workloads. The commit references a cross-link to the fbtriton repository, signaling tighter collaboration between PyTorch and Triton communities.