Small comment/doc fixes

docs/docker.md

@@ -185,7 +185,7 @@ The whole goal of building these kernels is to allow researchers, developers, an
 To load a kernel locally, you should add the kernel build that is compatible with the Torch and CUDA versions in your environment to `PYTHONPATH`. For example:
 
 ```bash
-# PyTorch 2.6 and CUDA 12.6
+# PyTorch 2.9 and CUDA 12.6
 export PYTHONPATH="result/torch29-cxx11-cu126-x86_64-linux"
 ```
 

tests/Dockerfile.test-kernel

@@ -1,6 +1,6 @@
 # syntax=docker/dockerfile:1.4
 ARG PYTHON_VERSION=3.10
-# Ideally we'd test with 11.8, but the GELU kernel is subtly off.
+# Test with the oldest CUDA version supported by the latest PyTorch.
 ARG CUDA_VERSION=12.6.0
 ARG UBI_VERSION=8
 ARG TORCH_VERSION=2.9.0