5090 compile error

[heapsoftware]

Running:
git clone https://github.com/wutzebaer/tensorflow-5090.git ; cd tensorflow-5090 ; docker build --progress=plain .

Got the following error:
456.3 [12,456 / 35,428] Compiling xla/backends/cpu/runtime/convolution_thunk_f32.cc; 45s local ... (13 actions, 11 running)
457.3 [12,484 / 35,428] Compiling xla/backends/cpu/runtime/convolution_thunk_f32.cc; 46s local ... (12 actions, 11 running)
458.3 [12,509 / 35,428] Compiling src/cpu/x64/gemm/f32/jit_avx512_core_f32_copy_at_kern_part2_autogen.cpp; 26s local ... (12 actions, 11 running)
459.1 ERROR: /root/.cache/bazel/_bazel_root/2c92b5569ddded7b3a6bd5e139451b60/external/local_xla/xla/service/gpu/kernels/BUILD:286:13: Compiling xla/service/gpu/kernels/cutlass_gemm_kernel_bf16xbf16_to_bf16.cu.cc failed: (Exit 1): clang failed: error executing CppCompile command (from target @@local_xla//xla/service/gpu/kernels:cutlass_gemm_kernel_bf16xbf16_to_bf16) external/rules_ml_toolchain/cc/impls/cuda/wrappers/linux/clang -MD -MF ... (remaining 226 arguments skipped)
459.1 clang: error: unsupported CUDA gpu architecture: sm_120
459.4 Target //tensorflow/tools/pip_package:wheel failed to build
459.4 Use --verbose_failures to see the command lines of failed build steps.
459.4 INFO: Elapsed time: 459.209s, Critical Path: 46.27s
459.4 INFO: 12555 processes: 6113 internal, 6442 local.
459.4 ERROR: Build did NOT complete successfully

Dockerfile:35

34 | # Build TensorFlow wheel
35 | >>> RUN --mount=type=cache,target=/root/.cache/bazel,id=bazel-cache
36 | >>> bazel build //tensorflow/tools/pip_package:wheel --repo_env=USE_PYWRAP_RULES=1 --repo_env=WHEEL_NAME=tensorflow --config=cuda --config=cuda_wheel &&
37 | >>> cp /workspace/tensorflow/bazel-bin/tensorflow/tools/pip_package/wheel_house/*.whl /workspace
38 |

ERROR: failed to build: failed to solve: process "/bin/sh -c bazel build //tensorflow/tools/pip_package:wheel --repo_env=USE_PYWRAP_RULES=1 --repo_env=WHEEL_NAME=tensorflow --config=cuda --config=cuda_wheel && cp /workspace/tensorflow/bazel-bin/tensorflow/tools/pip_package/wheel_house/*.whl /workspace" did not complete successfully: exit code: 1

[heapsoftware]

Looks like clang didnt get support for 120 until recently, still not sure why im getting this error when it looks like 20 should support it, any assist?
llvm/llvm-project#127187

[heapsoftware]

Couldnt compile myself on my 5090 for some reason, but when I did the following it is working, so will close for now, thanks for building this...
docker pull wutzebaer/tensorflow-5090:latest
docker run -it --rm --gpus all wutzebaer/tensorflow-5090 $(docker build -q .)

would like to know how to fix this to compile myself so I can add some extra pip packages into the docker file on build