public_git/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/whisper.cpp.git synced 2026-09-30 11:36:38 +02:00
Code Issues Packages Projects Releases Wiki Activity

devops : reduce Vulkan Docker image to 39.4% of its original size (now 692 MB) (#4038)

Browse Source 
* devops : do not install recommended packages

Also:
- Replace apt-get with apt
- Explicitly install ca-certificates

* devops : remove development packages from runtime

* devops : copy only CLI executables and shared libraries

Won't copy compiled tests, source code and other build artifacts.

* devops : remove bash as entrypoint

With this change, arguments passed to `docker run` are forwarded
directly to the selected command.

Having `bash -c` in the entrypoint will cause additional arguments to be
interpreted according to shell command-string semantics, which makes
commands such as model downloaders and CLI tools behave unexpectedly
unless `--entrypoint` is specified explicitly or the whole command is
quoted.

Removing it restores the conventional Docker behavior of: `docker run
IMAGE COMMAND ARG...` passing `ARG` to `COMMAND` (not to `bash`).

* docs : fix + enhance Docker examples
This commit is contained in:
Álvaro Justen
2026-09-21 08:19:14 +02:00
committed by GitHub
parent 5670d5c0bb
commit 307869af28
2 changed files with 59 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+40 -17
View File
@@ -586,31 +586,54 @@ We have multiple Docker images available for this project:
### Usage
```shell
# Use the main tag or: cublas, main-cuda, main-intel, main-musa, main-rocm, main-vulkan.
IMAGE="ghcr.io/ggml-org/whisper.cpp:main"
MODEL_PATH="/tmp/whisper.cpp-models"
AUDIO_PATH="/tmp/whisper.cpp-audio"
AUDIO_URL="https://github.com/ggml-org/whisper.cpp/raw/refs/heads/master/samples/jfk.wav"
mkdir -p "$MODEL_PATH" "$AUDIO_PATH"
wget -O "$AUDIO_PATH/jfk.wav" "$AUDIO_URL"
# download model and persist it in a local folder
docker run -it --rm \
-v path/to/models:/models \
whisper.cpp:main "./models/download-ggml-model.sh base /models"
-v $MODEL_PATH:/models \
$IMAGE \
download-ggml-model.sh base /models
# transcribe an audio file
docker run -it --rm \
-v path/to/models:/models \
-v path/to/audios:/audios \
whisper.cpp:main "whisper-cli -m /models/ggml-base.bin -f /audios/jfk.wav"
# transcribe an audio file in samples folder
docker run -it --rm \
-v path/to/models:/models \
whisper.cpp:main "whisper-cli -m /models/ggml-base.bin -f ./samples/jfk.wav"
-v $MODEL_PATH:/models \
-v $AUDIO_PATH:/audios \
$IMAGE \
whisper-cli -m /models/ggml-base.bin -f /audios/jfk.wav
# run the web server
docker run -it --rm -p "8080:8080" \
-v path/to/models:/models \
whisper.cpp:main "whisper-server --host 127.0.0.1 -m /models/ggml-base.bin"
# run the bench too on the small.en model using 4 threads
docker run -it --rm \
-v path/to/models:/models \
whisper.cpp:main "whisper-bench -m /models/ggml-small.en.bin -t 4"
-p "8080:8080" \
-v $MODEL_PATH:/models \
$IMAGE \
whisper-server --host 0.0.0.0 -m /models/ggml-base.bin
# then:
curl -v http://127.0.0.1:8080/inference \
-F "file=@${AUDIO_PATH}/jfk.wav" \
-F 'response_format=json'
# download small.en and run the bench on it using 4 threads
docker run -it --rm \
-v $MODEL_PATH:/models \
$IMAGE \
download-ggml-model.sh small.en /models
docker run -it --rm \
-v $MODEL_PATH:/models \
$IMAGE \
whisper-bench -m /models/ggml-small.en.bin -t 4
# the methods above use the CPU - use your GPU by sharing the device, for example, for an AMD iGPU via Vulkan:
docker run --rm \
--device /dev/dri \
-v $MODEL_PATH:/models \
"ghcr.io/ggml-org/whisper.cpp:main-vulkan" \
whisper-bench -m /models/ggml-small.en.bin -t 4
```
## Installing with Conan