Add arm build, bump to CUDA 13

[thespad]

---

• I have read the contributing guideline and understand that I have made the correct modifications

---

Description:

Closes #61

Benefits of this PR and context:

How Has This Been Tested?

Source / References:

[LinuxServer-CI]

I am a bot, here is the pushed image/manifest for this PR:

ghcr.io/linuxserver/lspipepr-faster-whisper:gpu-v3.1.0-pkg-fe08b6cf-dev-3f8a1092ee58cb863b101742e805a0fb66eb45de-pr-62

[LinuxServer-CI]

I am a bot, here is the pushed image/manifest for this PR:

ghcr.io/linuxserver/lspipepr-faster-whisper:gpu-v3.1.0-pkg-fe08b6cf-dev-c36aca5be3c22f67309f458bae735fedf2b9137b-pr-62

[osos]

I get this in the logs:

faster-whisper  | 2026-03-13T08:15:21.373206427Z ModuleNotFoundError: No module named 'nvidia.cublas'
faster-whisper  | 2026-03-13T08:15:21.707975920Z INFO:httpx:HTTP Request: GET https://huggingface.co/api/models/mobiuslabsgmbh/faster-whisper-large-v3-turbo/revision/main "HTTP/1.1 307 Temporary Redirect"
faster-whisper  | 2026-03-13T08:15:21.823524916Z INFO:httpx:HTTP Request: GET https://huggingface.co/api/models/dropbox-dash/faster-whisper-large-v3-turbo/revision/main "HTTP/1.1 200 OK"
faster-whisper  | 2026-03-13T08:15:21.828411127Z Traceback (most recent call last):
faster-whisper  | 2026-03-13T08:15:21.828515944Z   File "<frozen runpy>", line 198, in _run_module_as_main
faster-whisper  | 2026-03-13T08:15:21.828517640Z   File "<frozen runpy>", line 88, in _run_code
faster-whisper  | 2026-03-13T08:15:21.828518744Z   File "/lsiopy/lib/python3.12/site-packages/wyoming_faster_whisper/__main__.py", line 254, in <module>
faster-whisper  | 2026-03-13T08:15:21.828519992Z     run()
faster-whisper  | 2026-03-13T08:15:21.828521112Z   File "/lsiopy/lib/python3.12/site-packages/wyoming_faster_whisper/__main__.py", line 249, in run
faster-whisper  | 2026-03-13T08:15:21.828523832Z     asyncio.run(main())
faster-whisper  | 2026-03-13T08:15:21.828525096Z   File "/usr/lib/python3.12/asyncio/runners.py", line 194, in run
faster-whisper  | 2026-03-13T08:15:21.828566072Z     return runner.run(main)
faster-whisper  | 2026-03-13T08:15:21.828576008Z            ^^^^^^^^^^^^^^^^
faster-whisper  | 2026-03-13T08:15:21.828578648Z   File "/usr/lib/python3.12/asyncio/runners.py", line 118, in run
faster-whisper  | 2026-03-13T08:15:21.828580808Z     return self._loop.run_until_complete(task)
faster-whisper  | 2026-03-13T08:15:21.828582616Z            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
faster-whisper  | 2026-03-13T08:15:21.828584328Z   File "/usr/lib/python3.12/asyncio/base_events.py", line 687, in run_until_complete
faster-whisper  | 2026-03-13T08:15:21.828586328Z     return future.result()
faster-whisper  | 2026-03-13T08:15:21.828587976Z            ^^^^^^^^^^^^^^^
faster-whisper  | 2026-03-13T08:15:21.828589608Z   File "/lsiopy/lib/python3.12/site-packages/wyoming_faster_whisper/__main__.py", line 218, in main
faster-whisper  | 2026-03-13T08:15:21.828595656Z     await loader.load_transcriber()
faster-whisper  | 2026-03-13T08:15:21.828597432Z   File "/lsiopy/lib/python3.12/site-packages/wyoming_faster_whisper/models.py", line 154, in load_transcriber
faster-whisper  | 2026-03-13T08:15:21.828838824Z     transcriber = FasterWhisperTranscriber(
faster-whisper  | 2026-03-13T08:15:21.828851864Z                   ^^^^^^^^^^^^^^^^^^^^^^^^^
faster-whisper  | 2026-03-13T08:15:21.828854552Z   File "/lsiopy/lib/python3.12/site-packages/wyoming_faster_whisper/faster_whisper_handler.py", line 26, in __init__
faster-whisper  | 2026-03-13T08:15:21.828857768Z     self.model = faster_whisper.WhisperModel(
faster-whisper  | 2026-03-13T08:15:21.828859592Z                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
faster-whisper  | 2026-03-13T08:15:21.828861288Z   File "/lsiopy/lib/python3.12/site-packages/faster_whisper/transcribe.py", line 689, in __init__
faster-whisper  | 2026-03-13T08:15:21.828863144Z     self.model = ctranslate2.models.Whisper(
faster-whisper  | 2026-03-13T08:15:21.828864760Z                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^
faster-whisper  | 2026-03-13T08:15:21.828866312Z ValueError: This CTranslate2 package was not compiled with CUDA support

docker-compose.yml

  faster-whisper:
    #image: lscr.io/linuxserver/faster-whisper:latest
    image: ghcr.io/linuxserver/lspipepr-faster-whisper:gpu-v3.1.0-pkg-fe08b6cf-dev-c36aca5be3c22f67309f458bae735fedf2b9137b-pr-62
    container_name: faster-whisper
    environment:
      - PUID=1000
      - PGID=1000
      - DEBUG= #optional
      - LOCAL_ONLY=
      - WHISPER_BEAM=5 #optional
      - WHISPER_LANG=auto #optional
      - WHISPER_MODEL=large-v3-turbo #optional
      - WHISPER_DEVICE=cuda
    volumes:
      - ./faster-whisper/config:/config
    ports:
      - 10300:10300
    restart: unless-stopped
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

[thespad]

Ugh, looks like they don't build ctranslate2 for arm with cuda support and expect you to compile it from source if you want to.

[osos]

Found this: https://github.com/assix/ctranslate2-aarch64-cuda13-binaries
Not sure how to add or use.

[thespad]

Unfortunately we can't just ingest a random tarball of unknown provenance that may or may not ever be updated.

It may be possible to compile ctranslate2 from source with cuda support ourselves, but it is very poorly documented, so thus far while I can get it to build, I can't get the python wrapper to actually use the libs, and we don't have any arm hardware with a nvidia GPU to test with properly.

It doesn't help that it needs almost 10Gb of build deps.

[LinuxServer-CI]

I am a bot, here is the pushed image/manifest for this PR:

ghcr.io/linuxserver/lspipepr-faster-whisper:gpu-v3.1.0-pkg-a5c5b2a6-dev-15c20be00002a8c815c8270cb1f6463667a150c1-pr-62