From 01b24e75ea823361cb97094d4700f0ca362ee2ff Mon Sep 17 00:00:00 2001
From: streamer45 <cstcld91@gmail.com>
Date: Thu, 1 Jan 2026 10:36:38 +0100
Subject: [PATCH] chore: smaller demo image

---
 Dockerfile.demo       | 25 +++++++++++++------------
 docker-skit-demo.toml |  9 ++-------
 2 files changed, 15 insertions(+), 19 deletions(-)

diff --git a/Dockerfile.demo b/Dockerfile.demo
index 2b2732b9..c174b820 100644
--- a/Dockerfile.demo
+++ b/Dockerfile.demo
@@ -153,18 +153,12 @@ RUN --mount=type=cache,id=cargo-registry-whisper,target=/usr/local/cargo/registr
     mkdir -p /build/plugins/native && \
     cp target-portable/release/libwhisper.so /build/plugins/native/
 
-# Download Whisper models (quantized only - no unquantized ggml-base.en.bin)
-# - ggml-base.en-q5_1.bin: English STT (quantized)
-# - ggml-tiny.en-q5_1.bin: Lightweight English STT
-# - ggml-base-q5_1.bin: Multilingual (for ES->EN translation)
+# Download Whisper models (demo image uses a single tiny multilingual model).
+# - ggml-tiny-q5_1.bin: Tiny multilingual STT (quantized)
 # - silero_vad.onnx: VAD model for Whisper
 RUN mkdir -p /build/models && \
-    curl -L -o /build/models/ggml-base.en-q5_1.bin \
-      https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en-q5_1.bin && \
-    curl -L -o /build/models/ggml-tiny.en-q5_1.bin \
-      https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en-q5_1.bin && \
-    curl -L -o /build/models/ggml-base-q5_1.bin \
-      https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base-q5_1.bin && \
+    curl -L -o /build/models/ggml-tiny-q5_1.bin \
+      https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny-q5_1.bin && \
     curl -L -o /build/models/silero_vad.onnx \
       https://raw.githubusercontent.com/snakers4/silero-vad/master/src/silero_vad/data/silero_vad.onnx
 
@@ -498,8 +492,15 @@ COPY --chown=app:app samples/pipelines /opt/streamkit/samples/pipelines
 COPY --chown=app:app samples/audio/system/*.ogg samples/audio/system/*.ogg.license /opt/streamkit/samples/audio/system/
 COPY --chown=app:app samples/audio/system/*.opus samples/audio/system/*.opus.license /opt/streamkit/samples/audio/system/
 
-# Remove pipelines that won't work (NLLB + WASM dependencies not included)
-RUN rm -f \
+# Demo image uses a single tiny multilingual Whisper model, but the committed sample pipelines
+# reference a mix of base.en/tiny.en/multilingual model filenames. Rewrite them inside the image
+# so the shipped samples work without bundling multiple Whisper models.
+RUN find /opt/streamkit/samples/pipelines -type f \( -name '*.yml' -o -name '*.yaml' \) -print0 | \
+    xargs -0 sed -i \
+      -e 's#models/ggml-base\\.en-q5_1\\.bin#models/ggml-tiny-q5_1.bin#g' \
+      -e 's#models/ggml-tiny\\.en-q5_1\\.bin#models/ggml-tiny-q5_1.bin#g' \
+      -e 's#models/ggml-base-q5_1\\.bin#models/ggml-tiny-q5_1.bin#g' && \
+    rm -f \
     /opt/streamkit/samples/pipelines/oneshot/speech_to_text_translate.yml \
     /opt/streamkit/samples/pipelines/oneshot/gain_filter_rust.yml \
     /opt/streamkit/samples/pipelines/dynamic/speech-translate-en-es.yaml \
diff --git a/docker-skit-demo.toml b/docker-skit-demo.toml
index 76fd82b4..4182c224 100644
--- a/docker-skit-demo.toml
+++ b/docker-skit-demo.toml
@@ -44,15 +44,10 @@ keep_models_loaded = true
 [resources.prewarm]
 enabled = true
 
-# Whisper STT - base English model (used by most sample pipelines)
+# Whisper STT - demo image uses a single tiny multilingual model
 [[resources.prewarm.plugins]]
 kind = "plugin::native::whisper"
-params = { model_path = "models/ggml-base.en-q5_1.bin", use_gpu = false, gpu_device = 0, n_threads = 2, vad_model_path = "models/silero_vad.onnx" }
-
-# Whisper STT - multilingual model (used by speech-translate-helsinki-es-en.yaml)
-[[resources.prewarm.plugins]]
-kind = "plugin::native::whisper"
-params = { model_path = "models/ggml-base-q5_1.bin", use_gpu = false, gpu_device = 0, n_threads = 2, vad_model_path = "models/silero_vad.onnx" }
+params = { model_path = "models/ggml-tiny-q5_1.bin", use_gpu = false, gpu_device = 0, n_threads = 2, vad_model_path = "models/silero_vad.onnx" }
 
 # VAD (Voice Activity Detection)
 [[resources.prewarm.plugins]]