Add data parallel (DP) support to vLLM server launcher

- Add --dp/--data-parallel-size flag for running independent model replicas
  across multiple GPUs with automatic load balancing behind a single port
- Add --tp/--tensor-parallel-size flag (previously hardcoded to 1)
- Update docs/vibevoice-vllm-asr.md with multi-GPU deployment guide
  covering DP, TP, and hybrid (DP × TP) configurations

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

docs/vibevoice-vllm-asr.md

@@ -10,6 +10,7 @@ Deploy VibeVoice ASR model as a high-performance API service using [vLLM](https:
 - **📡 OpenAI-Compatible API**: Standard `/v1/chat/completions` endpoint with streaming support
 - **🎵 Long Audio Support**: Process up to 60+ minutes of audio in a single request
 - **🔌 Plugin Architecture**: No vLLM source code modification required - just install and run
+- **⚡ Data Parallel (DP)**: Run independent model replicas across multiple GPUs with automatic load balancing behind a single port
 
 ## 🛠️ Installation
 
@@ -35,6 +36,66 @@ docker run -d --gpus all --name vibevoice-vllm \
   -c "python3 /app/vllm_plugin/scripts/start_server.py"
 ```
 
+## ⚡ Multi-GPU Deployment
+
+The launcher supports two types of GPU parallelism via `--tp` and `--dp` flags:
+
+| Flag | Name | What it does |
+|------|------|-------------|
+| `--tp N` | Tensor Parallel | Splits **one model** across N GPUs (for models too large for a single GPU) |
+| `--dp N` | Data Parallel | Runs **N independent replicas**, one per GPU, with automatic load balancing behind a single port |
+
+### Data Parallel (Recommended for scaling throughput)
+
+Run 4 independent replicas on 4 GPUs — vLLM automatically distributes incoming requests:
+
+```bash
+docker run -d --gpus '"device=0,1,2,3"' --name vibevoice-vllm \
+  --ipc=host \
+  -p 8000:8000 \
+  -e VIBEVOICE_FFMPEG_MAX_CONCURRENCY=64 \
+  -e PYTORCH_ALLOC_CONF=expandable_segments:True \
+  -v $(pwd):/app \
+  -w /app \
+  --entrypoint bash \
+  vllm/vllm-openai:v0.14.1 \
+  -c "python3 /app/vllm_plugin/scripts/start_server.py --dp 4"
+```
+
+### Tensor Parallel
+
+Split a single model across 2 GPUs (useful if GPU memory is limited):
+
+```bash
+docker run -d --gpus '"device=0,1"' --name vibevoice-vllm \
+  --ipc=host \
+  -p 8000:8000 \
+  -e VIBEVOICE_FFMPEG_MAX_CONCURRENCY=64 \
+  -e PYTORCH_ALLOC_CONF=expandable_segments:True \
+  -v $(pwd):/app \
+  -w /app \
+  --entrypoint bash \
+  vllm/vllm-openai:v0.14.1 \
+  -c "python3 /app/vllm_plugin/scripts/start_server.py --tp 2"
+```
+
+### Hybrid (DP × TP)
+
+Combine both — e.g., 2 replicas, each split across 2 GPUs (4 GPUs total):
+
+```bash
+docker run -d --gpus '"device=0,1,2,3"' --name vibevoice-vllm \
+  --ipc=host \
+  -p 8000:8000 \
+  -v $(pwd):/app \
+  -w /app \
+  --entrypoint bash \
+  vllm/vllm-openai:v0.14.1 \
+  -c "python3 /app/vllm_plugin/scripts/start_server.py --dp 2 --tp 2"
+```
+
+> **Note**: Total GPUs required = `dp × tp`. Make sure to expose enough GPU devices in the Docker `--gpus` flag.
+
 3. View logs
 ```bash
 docker logs -f vibevoice-vllm

vllm_plugin/scripts/start_server.py

@@ -77,10 +77,14 @@ def generate_tokenizer(model_path: str) -> None:
     )
 
 
-def start_vllm_server(model_path: str, port: int) -> None:
+def start_vllm_server(model_path: str, port: int,
+                      tensor_parallel_size: int = 1,
+                      data_parallel_size: int = 1) -> None:
     """Start vLLM server (replaces current process)."""
     print(f"\n{'='*60}")
     print(f"  Starting vLLM server on port {port}")
+    print(f"  Tensor Parallel (TP): {tensor_parallel_size}")
+    print(f"  Data Parallel   (DP): {data_parallel_size}")
     print(f"{'='*60}\n")
     
     vllm_cmd = [
@@ -96,7 +100,8 @@ def start_vllm_server(model_path: str, port: int) -> None:
         "--no-enable-prefix-caching",
         "--enable-chunked-prefill",
         "--chat-template-content-format", "openai",
-        "--tensor-parallel-size", "1",
+        "--tensor-parallel-size", str(tensor_parallel_size),
+        "--data-parallel-size", str(data_parallel_size),
         "--allowed-local-media-path", "/app",
         "--port", str(port),
     ]
@@ -110,12 +115,18 @@ def main():
         formatter_class=argparse.RawDescriptionHelpFormatter,
         epilog="""
 Examples:
-    # Start with default settings
+    # Start with default settings (single GPU)
     python3 start_server.py
 
     # Use custom port
     python3 start_server.py --port 8080
 
+    # Data parallel: 4 independent replicas on 4 GPUs (load balancing)
+    python3 start_server.py --dp 4
+
+    # Tensor parallel: split model across 2 GPUs
+    python3 start_server.py --tp 2
+
     # Skip dependency installation (if already installed)
     python3 start_server.py --skip-deps
         """
@@ -141,6 +152,20 @@ Examples:
         action="store_true",
         help="Skip generating tokenizer files"
     )
+    parser.add_argument(
+        "--tp", "--tensor-parallel-size",
+        type=int,
+        default=1,
+        dest="tensor_parallel_size",
+        help="Tensor parallel size: split one model across N GPUs (default: 1)"
+    )
+    parser.add_argument(
+        "--dp", "--data-parallel-size",
+        type=int,
+        default=1,
+        dest="data_parallel_size",
+        help="Data parallel size: run N independent model replicas for load balancing (default: 1)"
+    )
     args = parser.parse_args()
 
     print("\n" + "="*60)
@@ -162,7 +187,9 @@ Examples:
         generate_tokenizer(model_path)
 
     # Step 5: Start vLLM server
-    start_vllm_server(model_path, args.port)
+    start_vllm_server(model_path, args.port,
+                      tensor_parallel_size=args.tensor_parallel_size,
+                      data_parallel_size=args.data_parallel_size)
 
 
 if __name__ == "__main__":