Misc. bug: Docker builds fail with 'No url found for submodule path 'ggml/src/ggml-kompute/kompute' in .gitmodules'

[sammcj]

Name and Version

• Latest HEAD (6efcd65)
• Official Cuda Dockerfile

Operating systems

Linux

Which llama.cpp modules do you know to be affected?

Other (Please specify in the next section)

Command line

docker compose --profile llama build llama

Problem description & steps to reproduce

When trying to build from the .devops/cuda.Dockerfile file builds fail due to a broken git submodule:

 => ERROR [internal] load git source https://github.com/ggml-org/llama.cpp.git                                                                                                           3.1s
------
 > [internal] load git source https://github.com/ggml-org/llama.cpp.git:
0.692 ref: refs/heads/master    HEAD
0.692 6efcd65945a98cf6883cdd9de4c8ccd8c79d219a  HEAD
1.375 6efcd65945a98cf6883cdd9de4c8ccd8c79d219a  refs/heads/master
0.712 ref: refs/heads/master    HEAD
0.712 6efcd65945a98cf6883cdd9de4c8ccd8c79d219a	HEAD
1.718 fatal: No url found for submodule path 'ggml/src/ggml-kompute/kompute' in .gitmodules
------
failed to solve: failed to read dockerfile: failed to update submodules for https://github.com/ggml-org/llama.cpp.git: git stderr:
fatal: No url found for submodule path 'ggml/src/ggml-kompute/kompute' in .gitmodules
: exit status 128

Docker build args (compose format):

services:
  &name llama:
    container_name: *name
    hostname: *name
    build:
      context: https://github.com/ggml-org/llama.cpp.git
      dockerfile: .devops/cuda.Dockerfile
      args:
        CUDA_DEVICE_ORDER: FASTEST
        CMAKE_BUILD_TYPE: Release
        CMAKE_CUDA_ARCHITECTURES: "8.6;8.6+PTX"
        CUDA_ARCHITECTURES: "8.6;8.6+PTX"
        CMAKE_EXE_LINKER_FLAGS: -Wl,--allow-shlib-undefined
        LLAMA_FATAL_WARNINGS: off
        GGML_CUDA_FA_ALL_QUANTS: on
        GGML_CUDA_USE_TENSOR_CORES: on
        GGML_CUDA_MMV_Y: 2
        GGML_CUDA_DMMV_X: 64
        GGML_CUDA_ENABLE_UNIFIED_MEMORY: on
        GGML_NATIVE: on
        GGML_CUDA: on
        GGML_AVX2: on
        GGML_AVX512: on
        GGML_AVX512_BF16: on
        GGML_SCHED_MAX_COPIES: 6
        GGML_AVX512_VBMI: on
        GGML_AVX512_VNNI: on
        GGML_AVX_VNNI: on
        GGML_SSE42: on
        GGML_CUDA_F16: on
        CUDA_F16: on
        GGML_CUDA_FA: on
        CUDA_FA: on
        GGML_CUDA_GRAPHS: on
        CUDA_GRAPHS: on
        GGML_CUDA_PEER_MAX_BATCH_SIZE: 512
        CUDA_USE_TENSOR_CORES: on
        PEER_MAX_BATCH_SIZE: 512
        CUDA_PEER_MAX_BATCH_SIZE: 512
        GGML_F16C: on
        GGML_LTO: on
        GGML_FMA: on
        LLAMA_BUILD_SERVER: on
        LLAMA_BUILD_TESTS: off
        LLAMA_SERVER_VERBOSE: off
        DEBIAN_FRONTEND: noninteractive

First Bad Commit

Unsure of the exact commit, but I believe it started having issues in just the past few days.

Relevant log output