Download vocab.json and merges.txt in container image

2 files changed, 5 insertions, 7 deletions

diff --git a/Dockerfile b/Dockerfile
index 4b6b432..48b74b6 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -22,11 +22,6 @@ ENV MODELS_DIR=/app/models
 ENV QWEN3_TTS_DIR=/app/models/qwen3-tts
 RUN /app/download-models.sh echo "Models downloaded"
 
-# Download missing Qwen3-TTS tokenizer files (vocab.json + merges.txt)
-# The main download script tries to get tokenizer.json which doesn't exist in the HF repo
-RUN curl -sL "https://huggingface.co/Qwen/Qwen3-TTS-12Hz-0.6B-Base/resolve/main/vocab.json" -o /app/models/qwen3-tts/vocab.json \
-    && curl -sL "https://huggingface.co/Qwen/Qwen3-TTS-12Hz-0.6B-Base/resolve/main/merges.txt" -o /app/models/qwen3-tts/merges.txt
-
 # Copy workspace files
 COPY Cargo.toml Cargo.lock ./
 COPY makima ./makima
diff --git a/makima/sh/download-models.sh b/makima/sh/download-models.sh
index e5e3689..4f188f3 100755
--- a/makima/sh/download-models.sh
+++ b/makima/sh/download-models.sh
@@ -121,7 +121,8 @@ download_qwen3_tts() {
     if [ -d "$QWEN3_TTS_DIR" ] && \
        [ -f "$QWEN3_TTS_DIR/model.safetensors" ] && \
        [ -f "$QWEN3_TTS_DIR/speech_tokenizer.safetensors" ] && \
-       [ -f "$QWEN3_TTS_DIR/tokenizer.json" ] && \
+       [ -f "$QWEN3_TTS_DIR/vocab.json" ] && \
+       [ -f "$QWEN3_TTS_DIR/merges.txt" ] && \
        [ -f "$QWEN3_TTS_DIR/config.json" ]; then
         echo "Qwen3-TTS models already exist, skipping..."
         return 0
@@ -131,11 +132,13 @@ download_qwen3_tts() {
     mkdir -p "$QWEN3_TTS_DIR"
 
     # Download base TTS model files from Qwen/Qwen3-TTS-12Hz-0.6B-Base
+    # Note: This repo uses vocab.json + merges.txt (not tokenizer.json)
     echo "Downloading Qwen3-TTS-12Hz-0.6B-Base..."
     hf download Qwen/Qwen3-TTS-12Hz-0.6B-Base \
         model.safetensors \
         config.json \
-        tokenizer.json \
+        vocab.json \
+        merges.txt \
         tokenizer_config.json \
         --local-dir "$QWEN3_TTS_DIR"