This is an umbrella ticked for LLVM SPIR-V backend enabling. As the backend matures and reaches non-experimental status, we can leverage it to eliminate the translator dependency. For the transition to run smoothly, we will stage the implementation alongside the translator approach until the backend reaches the same levels of functionality and performance. At its core, the transition for Triton will look like a localized change in how we produce SPIR-V from LLVM IR.