Use dustynv/vllm base image for Jetson CUDA support

Dockerfile

@@ -1,11 +1,9 @@
 # OrpheusTail - Orpheus TTS Service for NVIDIA Jetson AGX Orin
 # 
-# Replaces VoiceTail (Bark) with Orpheus for better emotion control
-# and voice cloning capabilities.
-#
-# Based on NVIDIA L4T PyTorch container optimized for Jetson
+# Uses dustynv's vLLM container which has proper Jetson CUDA support
+# Orpheus uses vLLM under the hood for fast inference
 
-FROM dustynv/pytorch:2.1-r36.2.0
+FROM dustynv/vllm:0.8.6-r36.4-cu128-24.04
 
 # Set working directory
 WORKDIR /app
@@ -20,13 +18,13 @@ RUN apt-get update && apt-get install -y \
 # Copy requirements first for better caching
 COPY requirements.txt /app/
 
-# Install Python dependencies
-# Note: torch and torchvision are already in the base image
+# Install Python dependencies (FastAPI, etc - but NOT torch/vllm)
 RUN pip3 install --no-cache-dir -r requirements.txt
 
-# Install orpheus-speech (uses vllm under the hood)
-# Note: vllm version compatibility may need adjustment
-RUN pip3 install orpheus-speech
+# Install orpheus-speech WITHOUT dependencies to avoid overwriting vllm/torch
+# Then install just the audio codec it needs (snac)
+RUN pip3 install --no-cache-dir --no-deps orpheus-speech && \
+    pip3 install --no-cache-dir snac
 
 # Copy application code
 COPY main.py /app/