Update start.sh

start.sh

@@ -1,69 +1,78 @@
 #!/bin/bash
-
-# Stop the script if any command fails
 set -e
 
 # --- CONFIGURATION ---
-# REMOVED SPACE AFTER =
-MODE="${MODEL_TYPE:-CODER}"
-
 echo "------------------------------------------------"
-echo "🛠️ Starting Space in [ $MODE ] mode"
+echo "🛠️  Setting up Multi-Model Environment"
 echo "------------------------------------------------"
 
-# Define Models - REMOVED SPACES AFTER =
-CODER_REPO="Qwen/Qwen2.5-Coder-14B-Instruct-GGUF"
-CODER_FILE="qwen2.5-coder-14b-instruct-q4_k_m.gguf"
-
-VISION_REPO="Qwen/Qwen2.5-VL-7B-Instruct-GGUF"
-VISION_FILE="qwen2.5-vl-7b-instruct-q4_k_m.gguf"
+# Create a dedicated directory for all models
+mkdir -p /app/models
 
-# Function to download
-download_if_needed() {
+# Function to download models
+download_model() {
     REPO_ID=$1
     FILENAME=$2
-    FILE_PATH="/app/$FILENAME"
+    # We rename it locally to make it easier to read in the UI
+    LOCAL_NAME=$3 
+
+    if [ -z "$LOCAL_NAME" ]; then
+        LOCAL_NAME="$FILENAME"
+    fi
+
+    FILE_PATH="/app/models/$LOCAL_NAME"
 
     if [ -f "$FILE_PATH" ]; then
-        echo "✅ Found existing model: $FILENAME"
+        echo "✅ Found existing model: $LOCAL_NAME"
     else
-        echo "⬇️ Model not found. Downloading $FILENAME ..."
+        echo "⬇️  Downloading $LOCAL_NAME from $REPO_ID ..."
         python3 -c "
 from huggingface_hub import hf_hub_download
 import sys
 try:
-    hf_hub_download(repo_id='$REPO_ID', filename='$FILENAME', local_dir='/app')
+    hf_hub_download(
+        repo_id='$REPO_ID',
+        filename='$FILENAME',
+        local_dir='/app/models',
+        local_dir_use_symlinks=False
+    )
+    # Rename if needed (hf_hub_download saves as original filename)
+    import os
+    if '$FILENAME' != '$LOCAL_NAME':
+        os.rename('/app/models/$FILENAME', '/app/models/$LOCAL_NAME')
     print('Download successful.')
 except Exception as e:
     print(f'Error: {e}')
-    sys.exit(1)
 "
     fi
 }
 
-# --- SELECTION LOGIC ---
-if [ "$MODE" = "VISION" ]; then
-    download_if_needed "$VISION_REPO" "$VISION_FILE"
-    MODEL_PATH="/app/$VISION_FILE"
-    CONTEXT=4096
-else
-    # Default to Coder
-    download_if_needed "$CODER_REPO" "$CODER_FILE"
-    MODEL_PATH="/app/$CODER_FILE"
-    CONTEXT=8192
-fi
+# 1. TEXT / CODE MODEL (Qwen 2.5 Coder 14B)
+download_model "Qwen/Qwen2.5-Coder-14B-Instruct-GGUF" \
+               "qwen2.5-coder-14b-instruct-q4_k_m.gguf"
+
+# 2. IMAGE / VISION MODEL (Qwen 2.5 VL 7B)
+# Note: Newer llama.cpp versions often bundle the projector, but sometimes separate mmproj is needed.
+# We will download the main model file.
+download_model "Qwen/Qwen2.5-VL-7B-Instruct-GGUF" \
+               "qwen2.5-vl-7b-instruct-q4_k_m.gguf"
+
+# 3. AUDIO MODEL (Qwen 2 Audio 7B)
+# Note: Support for Audio models in llama-server is experimental but available in newer builds
+download_model "Qwen/Qwen2-Audio-7B-Instruct-GGUF" \
+               "qwen2-audio-7b-instruct-q4_k_m.gguf"
+
 
 echo "------------------------------------------------"
-echo "🚀 Launching Llama.cpp Server..."
-echo "Model: $MODEL_PATH"
-echo "Context: $CONTEXT"
+echo "🚀 Launching Llama.cpp Server in Multi-Model Mode..."
 echo "------------------------------------------------"
 
-# Run Server
-# Updated to use absolute path to match your Dockerfile setup
+# We point --models-path to the folder. The server will auto-discover the GGUFs.
+# The UI should then allow you to pick which model to use.
 exec /usr/local/bin/llama-server \
-    -m "$MODEL_PATH" \
+    --model-url-path "" \
+    --models-path "/app/models" \
     --host 0.0.0.0 \
     --port 7860 \
-    -c $CONTEXT \
+    -c 8192 \
     --n-gpu-layers 0