Skip to content

fix(docker): move runtime defaults from CMD to env vars - #20

Merged
achetronic merged 2 commits into
masterfrom
fix/cuda-defaults-from-env
Jul 1, 2026
Merged

fix(docker): move runtime defaults from CMD to env vars#20
achetronic merged 2 commits into
masterfrom
fix/cuda-defaults-from-env

Conversation

@achetronic

Copy link
Copy Markdown
Owner

Docker replaces the whole CMD when the user passes their own args, so the defaults baked into CMD (-gpu cuda, -models /models, -port) were dropped on any override. The CUDA image silently ran CPU inference, and the models path fell back to ./models and crashed.

Fixes

  • Bake the defaults as PARAKEET_* env vars, which survive arg overrides. A flag passed on the command line still wins over the env var.
  • Applies to both the CPU and CUDA images.
  • Drop the redundant command block from the docker-compose example.

Relies on the generic flag-to-env mapping merged in #19.

Fixes #15.

Docker replaces the whole CMD when the user passes their own args, so the
defaults baked into CMD (-gpu cuda, -models /models, -port) were dropped on
any override: the CUDA image silently ran CPU inference and the models path
fell back to ./models and crashed.

Bake the defaults as PARAKEET_* env vars instead, which survive arg
overrides. A flag passed on the command line still wins over the env var.
Applies to both the CPU and CUDA images. Drops the redundant command block
from the docker-compose example.

Fixes #15.
@achetronic
achetronic force-pushed the fix/cuda-defaults-from-env branch from 672155b to 1db77c4 Compare July 1, 2026 13:40
@achetronic achetronic self-assigned this Jul 1, 2026
@achetronic achetronic added the enhancement New feature or request label Jul 1, 2026
@achetronic
achetronic merged commit e6965b3 into master Jul 1, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

CUDA image silently falls back to CPU when any CLI flag is passed

1 participant